prompt
stringlengths
94
42.6k
completion
stringlengths
6
120
api
stringlengths
14
68
# -*- coding: utf-8 -*- # MegEngine is Licensed under the Apache License, Version 2.0 (the "License") # # Copyright (c) 2014-2020 Megvii Inc. All rights reserved. # # Unless required by applicable law or agreed to in writing, # software distributed under the License is distributed on an # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. import argparse import json import os import subprocess import sys import time import numpy as np from resnet50 import Resnet50 import megengine as mge import megengine.distributed as dist import megengine.functional as F from megengine._internal.plugin import CompGraphProfiler from megengine.core import Graph, tensor from megengine.core.graph import get_default_graph from megengine.functional.debug_param import ( get_conv_execution_strategy, set_conv_execution_strategy, ) from megengine.jit import trace from megengine.module import BatchNorm2d, Conv2d, Linear, MaxPool2d, Module from megengine.optimizer import SGD sys.path.append(os.path.join(os.path.dirname(__file__), "..", "..", "..", "examples")) def init_profiler(comp_graph=get_default_graph()): profiler = CompGraphProfiler(comp_graph) return profiler def dump_profiler(profiler, filename): with open(filename, "w") as fout: json.dump(profiler.get(), fout, indent=2) def print_gpu_usage(): stdout = subprocess.getoutput("nvidia-smi") for line in stdout.split("\n"): for item in line.split(" "): if "MiB" in item: print("Finish with GPU Usage", item) break def run_perf( batch_size=64, warm_up=True, dump_prof=None, opt_level=2, conv_fastrun=False, run_step=True, track_bn_stats=True, warm_up_iter=20, run_iter=100, num_gpu=None, device=0, server=None, port=None, scale_batch_size=False, eager=False, ): if conv_fastrun:
set_conv_execution_strategy("PROFILE")
megengine.functional.debug_param.set_conv_execution_strategy
# -*- coding: utf-8 -*- # MegEngine is Licensed under the Apache License, Version 2.0 (the "License") # # Copyright (c) 2014-2020 Megvii Inc. All rights reserved. # # Unless required by applicable law or agreed to in writing, # software distributed under the License is distributed on an # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. import argparse import json import os import subprocess import sys import time import numpy as np from resnet50 import Resnet50 import megengine as mge import megengine.distributed as dist import megengine.functional as F from megengine._internal.plugin import CompGraphProfiler from megengine.core import Graph, tensor from megengine.core.graph import get_default_graph from megengine.functional.debug_param import ( get_conv_execution_strategy, set_conv_execution_strategy, ) from megengine.jit import trace from megengine.module import BatchNorm2d, Conv2d, Linear, MaxPool2d, Module from megengine.optimizer import SGD sys.path.append(os.path.join(os.path.dirname(__file__), "..", "..", "..", "examples")) def init_profiler(comp_graph=get_default_graph()): profiler = CompGraphProfiler(comp_graph) return profiler def dump_profiler(profiler, filename): with open(filename, "w") as fout: json.dump(profiler.get(), fout, indent=2) def print_gpu_usage(): stdout = subprocess.getoutput("nvidia-smi") for line in stdout.split("\n"): for item in line.split(" "): if "MiB" in item: print("Finish with GPU Usage", item) break def run_perf( batch_size=64, warm_up=True, dump_prof=None, opt_level=2, conv_fastrun=False, run_step=True, track_bn_stats=True, warm_up_iter=20, run_iter=100, num_gpu=None, device=0, server=None, port=None, scale_batch_size=False, eager=False, ): if conv_fastrun: set_conv_execution_strategy("PROFILE") if num_gpu:
dist.init_process_group(args.server, args.port, num_gpu, device, device)
megengine.distributed.init_process_group
# -*- coding: utf-8 -*- # MegEngine is Licensed under the Apache License, Version 2.0 (the "License") # # Copyright (c) 2014-2020 Megvii Inc. All rights reserved. # # Unless required by applicable law or agreed to in writing, # software distributed under the License is distributed on an # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. import argparse import json import os import subprocess import sys import time import numpy as np from resnet50 import Resnet50 import megengine as mge import megengine.distributed as dist import megengine.functional as F from megengine._internal.plugin import CompGraphProfiler from megengine.core import Graph, tensor from megengine.core.graph import get_default_graph from megengine.functional.debug_param import ( get_conv_execution_strategy, set_conv_execution_strategy, ) from megengine.jit import trace from megengine.module import BatchNorm2d, Conv2d, Linear, MaxPool2d, Module from megengine.optimizer import SGD sys.path.append(os.path.join(os.path.dirname(__file__), "..", "..", "..", "examples")) def init_profiler(comp_graph=get_default_graph()): profiler = CompGraphProfiler(comp_graph) return profiler def dump_profiler(profiler, filename): with open(filename, "w") as fout: json.dump(profiler.get(), fout, indent=2) def print_gpu_usage(): stdout = subprocess.getoutput("nvidia-smi") for line in stdout.split("\n"): for item in line.split(" "): if "MiB" in item: print("Finish with GPU Usage", item) break def run_perf( batch_size=64, warm_up=True, dump_prof=None, opt_level=2, conv_fastrun=False, run_step=True, track_bn_stats=True, warm_up_iter=20, run_iter=100, num_gpu=None, device=0, server=None, port=None, scale_batch_size=False, eager=False, ): if conv_fastrun: set_conv_execution_strategy("PROFILE") if num_gpu: dist.init_process_group(args.server, args.port, num_gpu, device, device) if scale_batch_size: batch_size = batch_size // num_gpu print("Run with data parallel, batch size = {} per GPU".format(batch_size)) data = tensor(np.random.randn(batch_size, 3, 224, 224).astype("float32")) label = tensor(np.random.randint(1000, size=[batch_size,], dtype=np.int32)) net = Resnet50(track_bn_stats=track_bn_stats) opt = SGD(net.parameters(), lr=0.01, momentum=0.9, weight_decay=1e-4) def train_func(data, label): logits = net(data) loss =
F.cross_entropy_with_softmax(logits, label)
megengine.functional.cross_entropy_with_softmax
# -*- coding: utf-8 -*- # MegEngine is Licensed under the Apache License, Version 2.0 (the "License") # # Copyright (c) 2014-2020 Megvii Inc. All rights reserved. # # Unless required by applicable law or agreed to in writing, # software distributed under the License is distributed on an # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. import argparse import json import os import subprocess import sys import time import numpy as np from resnet50 import Resnet50 import megengine as mge import megengine.distributed as dist import megengine.functional as F from megengine._internal.plugin import CompGraphProfiler from megengine.core import Graph, tensor from megengine.core.graph import get_default_graph from megengine.functional.debug_param import ( get_conv_execution_strategy, set_conv_execution_strategy, ) from megengine.jit import trace from megengine.module import BatchNorm2d, Conv2d, Linear, MaxPool2d, Module from megengine.optimizer import SGD sys.path.append(os.path.join(os.path.dirname(__file__), "..", "..", "..", "examples")) def init_profiler(comp_graph=get_default_graph()): profiler = CompGraphProfiler(comp_graph) return profiler def dump_profiler(profiler, filename): with open(filename, "w") as fout: json.dump(profiler.get(), fout, indent=2) def print_gpu_usage(): stdout = subprocess.getoutput("nvidia-smi") for line in stdout.split("\n"): for item in line.split(" "): if "MiB" in item: print("Finish with GPU Usage", item) break def run_perf( batch_size=64, warm_up=True, dump_prof=None, opt_level=2, conv_fastrun=False, run_step=True, track_bn_stats=True, warm_up_iter=20, run_iter=100, num_gpu=None, device=0, server=None, port=None, scale_batch_size=False, eager=False, ): if conv_fastrun: set_conv_execution_strategy("PROFILE") if num_gpu: dist.init_process_group(args.server, args.port, num_gpu, device, device) if scale_batch_size: batch_size = batch_size // num_gpu print("Run with data parallel, batch size = {} per GPU".format(batch_size)) data = tensor(np.random.randn(batch_size, 3, 224, 224).astype("float32")) label = tensor(np.random.randint(1000, size=[batch_size,], dtype=np.int32)) net = Resnet50(track_bn_stats=track_bn_stats) opt = SGD(net.parameters(), lr=0.01, momentum=0.9, weight_decay=1e-4) def train_func(data, label): logits = net(data) loss = F.cross_entropy_with_softmax(logits, label) if num_gpu: loss = loss / num_gpu opt.zero_grad() opt.backward(loss) return loss train_func = trace( train_func, symbolic=(not eager), opt_level=opt_level, profiling=not (dump_prof is None), ) if warm_up: print("Warm up ...") for _ in range(warm_up_iter): opt.zero_grad() train_func(data, label) if run_step: opt.step() print_gpu_usage() print("Running train ...") start = time.time() for _ in range(run_iter): opt.zero_grad() train_func(data, label) if run_step: opt.step() time_used = time.time() - start if dump_prof: with open(dump_prof, "w") as fout: json.dump(train_func.get_profile(), fout, indent=2) return time_used / run_iter def str2bool(v): if isinstance(v, bool): return v if v.lower() in ("yes", "true", "t", "y", "1"): return True elif v.lower() in ("no", "false", "f", "n", "0"): return False else: raise argparse.ArgumentTypeError("Boolean value expected.") if __name__ == "__main__": parser = argparse.ArgumentParser( description="Running regression test on Resnet 50", formatter_class=argparse.ArgumentDefaultsHelpFormatter, ) parser.add_argument("--batch-size", type=int, default=64, help="batch size ") parser.add_argument( "--warm-up", type=str2bool, default=True, help="whether to warm up" ) parser.add_argument( "--dump-prof", type=str, default=None, help="pass the json file path to dump the profiling result", ) parser.add_argument("--opt-level", type=int, default=2, help="graph opt level") parser.add_argument( "--conv-fastrun", type=str2bool, default=False, help="whether to use conv fastrun mode", ) parser.add_argument( "--run-step", type=str2bool, default=True, help="whether to run optimizer.step()", ) parser.add_argument( "--track-bn-stats", type=str2bool, default=True, help="whether to track bn stats", ) parser.add_argument( "--warm-up-iter", type=int, default=20, help="number of iters to warm up" ) parser.add_argument( "--run-iter", type=int, default=100, help="number of iters to collect wall time" ) parser.add_argument("--server", default="0.0.0.0") parser.add_argument("--port", type=int, default=2222) parser.add_argument( "--scale-batch-size", type=str2bool, default=False, help="whether to divide batch size by number of GPUs", ) parser.add_argument( "--eager", type=str2bool, default=False, help="whether to use eager mode" ) # Data parallel related parser.add_argument("--num-gpu", type=int, default=None) parser.add_argument("--device", type=int, default=0) args = parser.parse_args() print(vars(args)) os.environ["MGB_JIT_BACKEND"] = "NVRTC" t = run_perf(**vars(args)) print("**********************************") print("Wall time per iter {:.0f} ms".format(t * 1000)) print("**********************************")
get_default_graph()
megengine.core.graph.get_default_graph
# MegEngine is Licensed under the Apache License, Version 2.0 (the "License") # # Copyright (c) 2014-2020 Megvii Inc. All rights reserved. # # Unless required by applicable law or agreed to in writing, # software distributed under the License is distributed on an # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. import numpy as np import megengine._internal as mgb from ... import functional as F from ...core import Parameter from ..qat import linear as QAT from .module import QuantizedModule class Linear(QuantizedModule): r"""quantized version of :class:`~.qat.linear.Linear`.""" def __init__( self, dtype: np.dtype = None, ): super().__init__() self.weight = None self.bias = None self.output_dtype = dtype def forward(self, inp): if self.training: raise ValueError("quantized module only support inference.") inp_scale =
mgb.dtype.get_scale(inp.dtype)
megengine._internal.dtype.get_scale
# MegEngine is Licensed under the Apache License, Version 2.0 (the "License") # # Copyright (c) 2014-2020 Megvii Inc. All rights reserved. # # Unless required by applicable law or agreed to in writing, # software distributed under the License is distributed on an # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. import numpy as np import megengine._internal as mgb from ... import functional as F from ...core import Parameter from ..qat import linear as QAT from .module import QuantizedModule class Linear(QuantizedModule): r"""quantized version of :class:`~.qat.linear.Linear`.""" def __init__( self, dtype: np.dtype = None, ): super().__init__() self.weight = None self.bias = None self.output_dtype = dtype def forward(self, inp): if self.training: raise ValueError("quantized module only support inference.") inp_scale = mgb.dtype.get_scale(inp.dtype) w_scale =
mgb.dtype.get_scale(self.weight.dtype)
megengine._internal.dtype.get_scale
# MegEngine is Licensed under the Apache License, Version 2.0 (the "License") # # Copyright (c) 2014-2020 Megvii Inc. All rights reserved. # # Unless required by applicable law or agreed to in writing, # software distributed under the License is distributed on an # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. import numpy as np import megengine._internal as mgb from ... import functional as F from ...core import Parameter from ..qat import linear as QAT from .module import QuantizedModule class Linear(QuantizedModule): r"""quantized version of :class:`~.qat.linear.Linear`.""" def __init__( self, dtype: np.dtype = None, ): super().__init__() self.weight = None self.bias = None self.output_dtype = dtype def forward(self, inp): if self.training: raise ValueError("quantized module only support inference.") inp_scale = mgb.dtype.get_scale(inp.dtype) w_scale = mgb.dtype.get_scale(self.weight.dtype) bias_dtype =
mgb.dtype.qint32(inp_scale * w_scale)
megengine._internal.dtype.qint32
import os import math import argparse from multiprocessing import Process, Queue from tqdm import tqdm import numpy as np import megengine as mge from megengine import jit from config import config import network import dataset import misc_utils if_set_nms = True def eval_all(args): # model_path saveDir = config.model_dir evalDir = config.eval_dir misc_utils.ensure_dir(evalDir) model_file = os.path.join(saveDir, 'epoch_{}.pkl'.format(args.resume_weights)) assert os.path.exists(model_file) # load data records = misc_utils.load_json_lines(config.eval_source) # multiprocessing num_records = len(records) num_devs = args.devices num_image = math.ceil(num_records / num_devs) result_queue = Queue(1000) procs = [] all_results = [] for i in range(num_devs): start = i * num_image end = min(start + num_image, num_records) split_records = records[start:end] proc = Process(target=inference, args=( model_file, i, split_records, result_queue)) proc.start() procs.append(proc) pbar = tqdm(total=num_records, ncols=50) for i in range(num_records): t = result_queue.get() all_results.append(t) pbar.update(1) for p in procs: p.join() fpath = os.path.join(evalDir, 'dump-{}.json'.format(args.resume_weights)) misc_utils.save_json_lines(all_results, fpath) def inference(model_file, device, records, result_queue): @
jit.trace(symbolic=False)
megengine.jit.trace
import os import math import argparse from multiprocessing import Process, Queue from tqdm import tqdm import numpy as np import megengine as mge from megengine import jit from config import config import network import dataset import misc_utils if_set_nms = True def eval_all(args): # model_path saveDir = config.model_dir evalDir = config.eval_dir misc_utils.ensure_dir(evalDir) model_file = os.path.join(saveDir, 'epoch_{}.pkl'.format(args.resume_weights)) assert os.path.exists(model_file) # load data records = misc_utils.load_json_lines(config.eval_source) # multiprocessing num_records = len(records) num_devs = args.devices num_image = math.ceil(num_records / num_devs) result_queue = Queue(1000) procs = [] all_results = [] for i in range(num_devs): start = i * num_image end = min(start + num_image, num_records) split_records = records[start:end] proc = Process(target=inference, args=( model_file, i, split_records, result_queue)) proc.start() procs.append(proc) pbar = tqdm(total=num_records, ncols=50) for i in range(num_records): t = result_queue.get() all_results.append(t) pbar.update(1) for p in procs: p.join() fpath = os.path.join(evalDir, 'dump-{}.json'.format(args.resume_weights)) misc_utils.save_json_lines(all_results, fpath) def inference(model_file, device, records, result_queue): @jit.trace(symbolic=False) def val_func(): pred_boxes = net(net.inputs) return pred_boxes net = network.Network() net.eval() check_point =
mge.load(model_file)
megengine.load
import megengine as mge import megengine.module as M import pytest from basecls.models.snet import SNV2Block, SNV2XceptionBlock @pytest.mark.parametrize("w_in", [32, 48]) @pytest.mark.parametrize("w_out", [64]) @pytest.mark.parametrize("w_mid", [32, 24]) @pytest.mark.parametrize("stride", [1, 2]) @pytest.mark.parametrize("kernel", [3, 5]) @pytest.mark.parametrize("se_r", [0.0, 0.25]) @pytest.mark.parametrize("drop_path_prob", [0.0, 0.1]) @pytest.mark.parametrize("norm_name", ["BN"]) @pytest.mark.parametrize("act_name", ["relu"]) def test_block( w_in: int, w_out: int, w_mid: int, *, kernel: int, stride: int, norm_name: str, act_name: str, se_r: float, drop_path_prob: float, ): m = SNV2Block( w_in, w_out, w_mid, kernel=kernel, stride=stride, norm_name=norm_name, act_name=act_name, se_r=se_r, drop_path_prob=drop_path_prob, ) assert isinstance(m, M.Module) m(
mge.random.normal(size=(2, w_in * 2 // stride, 8, 8))
megengine.random.normal
import megengine as mge import megengine.module as M import pytest from basecls.models.snet import SNV2Block, SNV2XceptionBlock @pytest.mark.parametrize("w_in", [32, 48]) @pytest.mark.parametrize("w_out", [64]) @pytest.mark.parametrize("w_mid", [32, 24]) @pytest.mark.parametrize("stride", [1, 2]) @pytest.mark.parametrize("kernel", [3, 5]) @pytest.mark.parametrize("se_r", [0.0, 0.25]) @pytest.mark.parametrize("drop_path_prob", [0.0, 0.1]) @pytest.mark.parametrize("norm_name", ["BN"]) @pytest.mark.parametrize("act_name", ["relu"]) def test_block( w_in: int, w_out: int, w_mid: int, *, kernel: int, stride: int, norm_name: str, act_name: str, se_r: float, drop_path_prob: float, ): m = SNV2Block( w_in, w_out, w_mid, kernel=kernel, stride=stride, norm_name=norm_name, act_name=act_name, se_r=se_r, drop_path_prob=drop_path_prob, ) assert isinstance(m, M.Module) m(mge.random.normal(size=(2, w_in * 2 // stride, 8, 8))) @pytest.mark.parametrize("w_in", [32]) @pytest.mark.parametrize("w_out", [64]) @pytest.mark.parametrize("w_mid", [32]) @pytest.mark.parametrize("stride", [1, 2]) @pytest.mark.parametrize("kernel", [7, "x"]) @pytest.mark.parametrize("se_r", [0.25]) @pytest.mark.parametrize("drop_path_prob", [0.1]) @pytest.mark.parametrize("norm_name", ["BN"]) @pytest.mark.parametrize("act_name", ["relu"]) def test_x_block( w_in: int, w_out: int, w_mid: int, *, kernel: int, stride: int, norm_name: str, act_name: str, se_r: float, drop_path_prob: float, ): m = SNV2XceptionBlock( w_in, w_out, w_mid, kernel=kernel, stride=stride, norm_name=norm_name, act_name=act_name, se_r=se_r, drop_path_prob=drop_path_prob, ) assert isinstance(m, M.Module) m(
mge.random.normal(size=(2, w_in * 2 // stride, 8, 8))
megengine.random.normal
#!/usr/bin/env python3 # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. import megengine as mge import megengine.module as M import numpy as np import pytest import torch import torch.nn as nn from basecls.configs import BaseConfig from basecls.layers import BinaryCrossEntropy, CrossEntropy, build_loss @pytest.mark.parametrize("name", [CrossEntropy, "BinaryCrossEntropy", "CrossEntropy"]) def test_build_loss(name): cfg = BaseConfig(loss=dict(name=name)) m = build_loss(cfg) assert isinstance(m, M.Module) def test_bce(): x = np.random.rand(2, 8, 4).astype("float32") y = np.random.rand(2, 8, 4).astype("float32") ml = BinaryCrossEntropy()(
mge.Tensor(x)
megengine.Tensor
#!/usr/bin/env python3 # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. import megengine as mge import megengine.module as M import numpy as np import pytest import torch import torch.nn as nn from basecls.configs import BaseConfig from basecls.layers import BinaryCrossEntropy, CrossEntropy, build_loss @pytest.mark.parametrize("name", [CrossEntropy, "BinaryCrossEntropy", "CrossEntropy"]) def test_build_loss(name): cfg = BaseConfig(loss=dict(name=name)) m = build_loss(cfg) assert isinstance(m, M.Module) def test_bce(): x = np.random.rand(2, 8, 4).astype("float32") y = np.random.rand(2, 8, 4).astype("float32") ml = BinaryCrossEntropy()(mge.Tensor(x),
mge.Tensor(y)
megengine.Tensor
#!/usr/bin/env python3 # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. import megengine as mge import megengine.module as M import numpy as np import pytest import torch import torch.nn as nn from basecls.configs import BaseConfig from basecls.layers import BinaryCrossEntropy, CrossEntropy, build_loss @pytest.mark.parametrize("name", [CrossEntropy, "BinaryCrossEntropy", "CrossEntropy"]) def test_build_loss(name): cfg = BaseConfig(loss=dict(name=name)) m = build_loss(cfg) assert isinstance(m, M.Module) def test_bce(): x = np.random.rand(2, 8, 4).astype("float32") y = np.random.rand(2, 8, 4).astype("float32") ml = BinaryCrossEntropy()(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.BCEWithLogitsLoss()(torch.tensor(x), torch.tensor(y)).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) def test_ce(): K = 4 x = np.random.rand(2, 8, K).astype("float32") y = np.random.randint(K, size=(2, 8)).astype("int32") oy = np.eye(K, dtype="int32")[y] ml = CrossEntropy(axis=2)(
mge.Tensor(x)
megengine.Tensor
#!/usr/bin/env python3 # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. import megengine as mge import megengine.module as M import numpy as np import pytest import torch import torch.nn as nn from basecls.configs import BaseConfig from basecls.layers import BinaryCrossEntropy, CrossEntropy, build_loss @pytest.mark.parametrize("name", [CrossEntropy, "BinaryCrossEntropy", "CrossEntropy"]) def test_build_loss(name): cfg = BaseConfig(loss=dict(name=name)) m = build_loss(cfg) assert isinstance(m, M.Module) def test_bce(): x = np.random.rand(2, 8, 4).astype("float32") y = np.random.rand(2, 8, 4).astype("float32") ml = BinaryCrossEntropy()(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.BCEWithLogitsLoss()(torch.tensor(x), torch.tensor(y)).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) def test_ce(): K = 4 x = np.random.rand(2, 8, K).astype("float32") y = np.random.randint(K, size=(2, 8)).astype("int32") oy = np.eye(K, dtype="int32")[y] ml = CrossEntropy(axis=2)(mge.Tensor(x),
mge.Tensor(y)
megengine.Tensor
#!/usr/bin/env python3 # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. import megengine as mge import megengine.module as M import numpy as np import pytest import torch import torch.nn as nn from basecls.configs import BaseConfig from basecls.layers import BinaryCrossEntropy, CrossEntropy, build_loss @pytest.mark.parametrize("name", [CrossEntropy, "BinaryCrossEntropy", "CrossEntropy"]) def test_build_loss(name): cfg = BaseConfig(loss=dict(name=name)) m = build_loss(cfg) assert isinstance(m, M.Module) def test_bce(): x = np.random.rand(2, 8, 4).astype("float32") y = np.random.rand(2, 8, 4).astype("float32") ml = BinaryCrossEntropy()(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.BCEWithLogitsLoss()(torch.tensor(x), torch.tensor(y)).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) def test_ce(): K = 4 x = np.random.rand(2, 8, K).astype("float32") y = np.random.randint(K, size=(2, 8)).astype("int32") oy = np.eye(K, dtype="int32")[y] ml = CrossEntropy(axis=2)(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.CrossEntropyLoss()( torch.tensor(x).reshape(-1, K), torch.tensor(y).flatten().long() ).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) # one hot ol = CrossEntropy(axis=2)(
mge.Tensor(x)
megengine.Tensor
#!/usr/bin/env python3 # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. import megengine as mge import megengine.module as M import numpy as np import pytest import torch import torch.nn as nn from basecls.configs import BaseConfig from basecls.layers import BinaryCrossEntropy, CrossEntropy, build_loss @pytest.mark.parametrize("name", [CrossEntropy, "BinaryCrossEntropy", "CrossEntropy"]) def test_build_loss(name): cfg = BaseConfig(loss=dict(name=name)) m = build_loss(cfg) assert isinstance(m, M.Module) def test_bce(): x = np.random.rand(2, 8, 4).astype("float32") y = np.random.rand(2, 8, 4).astype("float32") ml = BinaryCrossEntropy()(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.BCEWithLogitsLoss()(torch.tensor(x), torch.tensor(y)).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) def test_ce(): K = 4 x = np.random.rand(2, 8, K).astype("float32") y = np.random.randint(K, size=(2, 8)).astype("int32") oy = np.eye(K, dtype="int32")[y] ml = CrossEntropy(axis=2)(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.CrossEntropyLoss()( torch.tensor(x).reshape(-1, K), torch.tensor(y).flatten().long() ).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) # one hot ol = CrossEntropy(axis=2)(mge.Tensor(x),
mge.Tensor(oy)
megengine.Tensor
#!/usr/bin/env python3 # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. import megengine as mge import megengine.module as M import numpy as np import pytest import torch import torch.nn as nn from basecls.configs import BaseConfig from basecls.layers import BinaryCrossEntropy, CrossEntropy, build_loss @pytest.mark.parametrize("name", [CrossEntropy, "BinaryCrossEntropy", "CrossEntropy"]) def test_build_loss(name): cfg = BaseConfig(loss=dict(name=name)) m = build_loss(cfg) assert isinstance(m, M.Module) def test_bce(): x = np.random.rand(2, 8, 4).astype("float32") y = np.random.rand(2, 8, 4).astype("float32") ml = BinaryCrossEntropy()(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.BCEWithLogitsLoss()(torch.tensor(x), torch.tensor(y)).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) def test_ce(): K = 4 x = np.random.rand(2, 8, K).astype("float32") y = np.random.randint(K, size=(2, 8)).astype("int32") oy = np.eye(K, dtype="int32")[y] ml = CrossEntropy(axis=2)(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.CrossEntropyLoss()( torch.tensor(x).reshape(-1, K), torch.tensor(y).flatten().long() ).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) # one hot ol = CrossEntropy(axis=2)(mge.Tensor(x), mge.Tensor(oy)).numpy() np.testing.assert_allclose(ml, ol, rtol=1e-4, atol=1e-6) # label smoothing ml = CrossEntropy(axis=2, label_smooth=0.1)(
mge.Tensor(x)
megengine.Tensor
#!/usr/bin/env python3 # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. import megengine as mge import megengine.module as M import numpy as np import pytest import torch import torch.nn as nn from basecls.configs import BaseConfig from basecls.layers import BinaryCrossEntropy, CrossEntropy, build_loss @pytest.mark.parametrize("name", [CrossEntropy, "BinaryCrossEntropy", "CrossEntropy"]) def test_build_loss(name): cfg = BaseConfig(loss=dict(name=name)) m = build_loss(cfg) assert isinstance(m, M.Module) def test_bce(): x = np.random.rand(2, 8, 4).astype("float32") y = np.random.rand(2, 8, 4).astype("float32") ml = BinaryCrossEntropy()(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.BCEWithLogitsLoss()(torch.tensor(x), torch.tensor(y)).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) def test_ce(): K = 4 x = np.random.rand(2, 8, K).astype("float32") y = np.random.randint(K, size=(2, 8)).astype("int32") oy = np.eye(K, dtype="int32")[y] ml = CrossEntropy(axis=2)(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.CrossEntropyLoss()( torch.tensor(x).reshape(-1, K), torch.tensor(y).flatten().long() ).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) # one hot ol = CrossEntropy(axis=2)(mge.Tensor(x), mge.Tensor(oy)).numpy() np.testing.assert_allclose(ml, ol, rtol=1e-4, atol=1e-6) # label smoothing ml = CrossEntropy(axis=2, label_smooth=0.1)(mge.Tensor(x),
mge.Tensor(y)
megengine.Tensor
#!/usr/bin/env python3 # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. import megengine as mge import megengine.module as M import numpy as np import pytest import torch import torch.nn as nn from basecls.configs import BaseConfig from basecls.layers import BinaryCrossEntropy, CrossEntropy, build_loss @pytest.mark.parametrize("name", [CrossEntropy, "BinaryCrossEntropy", "CrossEntropy"]) def test_build_loss(name): cfg = BaseConfig(loss=dict(name=name)) m = build_loss(cfg) assert isinstance(m, M.Module) def test_bce(): x = np.random.rand(2, 8, 4).astype("float32") y = np.random.rand(2, 8, 4).astype("float32") ml = BinaryCrossEntropy()(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.BCEWithLogitsLoss()(torch.tensor(x), torch.tensor(y)).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) def test_ce(): K = 4 x = np.random.rand(2, 8, K).astype("float32") y = np.random.randint(K, size=(2, 8)).astype("int32") oy = np.eye(K, dtype="int32")[y] ml = CrossEntropy(axis=2)(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.CrossEntropyLoss()( torch.tensor(x).reshape(-1, K), torch.tensor(y).flatten().long() ).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) # one hot ol = CrossEntropy(axis=2)(mge.Tensor(x), mge.Tensor(oy)).numpy() np.testing.assert_allclose(ml, ol, rtol=1e-4, atol=1e-6) # label smoothing ml = CrossEntropy(axis=2, label_smooth=0.1)(mge.Tensor(x), mge.Tensor(y)).numpy() ol = CrossEntropy(axis=2, label_smooth=0.1)(
mge.Tensor(x)
megengine.Tensor
#!/usr/bin/env python3 # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. import megengine as mge import megengine.module as M import numpy as np import pytest import torch import torch.nn as nn from basecls.configs import BaseConfig from basecls.layers import BinaryCrossEntropy, CrossEntropy, build_loss @pytest.mark.parametrize("name", [CrossEntropy, "BinaryCrossEntropy", "CrossEntropy"]) def test_build_loss(name): cfg = BaseConfig(loss=dict(name=name)) m = build_loss(cfg) assert isinstance(m, M.Module) def test_bce(): x = np.random.rand(2, 8, 4).astype("float32") y = np.random.rand(2, 8, 4).astype("float32") ml = BinaryCrossEntropy()(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.BCEWithLogitsLoss()(torch.tensor(x), torch.tensor(y)).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) def test_ce(): K = 4 x = np.random.rand(2, 8, K).astype("float32") y = np.random.randint(K, size=(2, 8)).astype("int32") oy = np.eye(K, dtype="int32")[y] ml = CrossEntropy(axis=2)(mge.Tensor(x), mge.Tensor(y)).numpy() tl = nn.CrossEntropyLoss()( torch.tensor(x).reshape(-1, K), torch.tensor(y).flatten().long() ).numpy() np.testing.assert_allclose(ml, tl, rtol=1e-4, atol=1e-6) # one hot ol = CrossEntropy(axis=2)(mge.Tensor(x), mge.Tensor(oy)).numpy() np.testing.assert_allclose(ml, ol, rtol=1e-4, atol=1e-6) # label smoothing ml = CrossEntropy(axis=2, label_smooth=0.1)(mge.Tensor(x), mge.Tensor(y)).numpy() ol = CrossEntropy(axis=2, label_smooth=0.1)(mge.Tensor(x),
mge.Tensor(oy)
megengine.Tensor
import io import numpy as np import megengine.core.tensor.megbrain_graph as G import megengine.utils.comp_graph_tools as cgtools from megengine import tensor from megengine.jit import trace from megengine.utils.network_node import VarNode def _default_compare_fn(x, y): if isinstance(x, np.ndarray): np.testing.assert_allclose(x, y, rtol=1e-6) else: np.testing.assert_allclose(x.numpy(), y, rtol=1e-6) def make_tensor(x, network=None, device=None): if network is not None: if isinstance(x, VarNode): return VarNode(x.var) return network.make_const(x, device=device) else: return
tensor(x, device=device)
megengine.tensor
import io import numpy as np import megengine.core.tensor.megbrain_graph as G import megengine.utils.comp_graph_tools as cgtools from megengine import tensor from megengine.jit import trace from megengine.utils.network_node import VarNode def _default_compare_fn(x, y): if isinstance(x, np.ndarray): np.testing.assert_allclose(x, y, rtol=1e-6) else: np.testing.assert_allclose(x.numpy(), y, rtol=1e-6) def make_tensor(x, network=None, device=None): if network is not None: if isinstance(x, VarNode): return VarNode(x.var) return network.make_const(x, device=device) else: return tensor(x, device=device) def opr_test( cases, func, compare_fn=_default_compare_fn, ref_fn=None, test_trace=True, network=None, **kwargs ): """ :param cases: the list which have dict element, the list length should be 2 for dynamic shape test. and the dict should have input, and should have output if ref_fn is None. should use list for multiple inputs and outputs for each case. :param func: the function to run opr. :param compare_fn: the function to compare the result and expected, use ``np.testing.assert_allclose`` if None. :param ref_fn: the function to generate expected data, should assign output if None. Examples: .. code-block:: dtype = np.float32 cases = [{"input": [10, 20]}, {"input": [20, 30]}] opr_test(cases, F.eye, ref_fn=lambda n, m: np.eye(n, m).astype(dtype), dtype=dtype) """ def check_results(results, expected): if not isinstance(results, (tuple, list)): results = (results,) for r, e in zip(results, expected): if not isinstance(r, (tensor, VarNode)): r = tensor(r) compare_fn(r, e) def get_param(cases, idx): case = cases[idx] inp = case.get("input", None) outp = case.get("output", None) if inp is None: raise ValueError("the test case should have input") if not isinstance(inp, (tuple, list)): inp = (inp,) if ref_fn is not None and callable(ref_fn): outp = ref_fn(*inp) if outp is None: raise ValueError("the test case should have output or reference function") if not isinstance(outp, (tuple, list)): outp = (outp,) return inp, outp if len(cases) == 0: raise ValueError("should give one case at least") if not callable(func): raise ValueError("the input func should be callable") inp, outp = get_param(cases, 0) inp_tensor = [make_tensor(inpi, network) for inpi in inp] if test_trace and not network: copied_inp = inp_tensor.copy() for symbolic in [False, True]: traced_func = trace(symbolic=symbolic)(func) for _ in range(3): traced_results = traced_func(*copied_inp, **kwargs) check_results(traced_results, outp) dumped_func = trace(symbolic=True, capture_as_const=True)(func) dumped_results = dumped_func(*copied_inp, **kwargs) check_results(dumped_results, outp) file = io.BytesIO() dump_info = dumped_func.dump(file) file.seek(0) # arg_name has pattern arg_xxx, xxx is int value def take_number(arg_name): return int(arg_name.split("_")[-1]) input_names = dump_info[4] inps_np = [i.numpy() for i in copied_inp] input_names.sort(key=take_number) inp_dict = dict(zip(input_names, inps_np)) infer_cg =
cgtools.GraphInference(file)
megengine.utils.comp_graph_tools.GraphInference
import io import numpy as np import megengine.core.tensor.megbrain_graph as G import megengine.utils.comp_graph_tools as cgtools from megengine import tensor from megengine.jit import trace from megengine.utils.network_node import VarNode def _default_compare_fn(x, y): if isinstance(x, np.ndarray): np.testing.assert_allclose(x, y, rtol=1e-6) else: np.testing.assert_allclose(x.numpy(), y, rtol=1e-6) def make_tensor(x, network=None, device=None): if network is not None: if isinstance(x, VarNode): return
VarNode(x.var)
megengine.utils.network_node.VarNode
import io import numpy as np import megengine.core.tensor.megbrain_graph as G import megengine.utils.comp_graph_tools as cgtools from megengine import tensor from megengine.jit import trace from megengine.utils.network_node import VarNode def _default_compare_fn(x, y): if isinstance(x, np.ndarray): np.testing.assert_allclose(x, y, rtol=1e-6) else: np.testing.assert_allclose(x.numpy(), y, rtol=1e-6) def make_tensor(x, network=None, device=None): if network is not None: if isinstance(x, VarNode): return VarNode(x.var) return network.make_const(x, device=device) else: return tensor(x, device=device) def opr_test( cases, func, compare_fn=_default_compare_fn, ref_fn=None, test_trace=True, network=None, **kwargs ): """ :param cases: the list which have dict element, the list length should be 2 for dynamic shape test. and the dict should have input, and should have output if ref_fn is None. should use list for multiple inputs and outputs for each case. :param func: the function to run opr. :param compare_fn: the function to compare the result and expected, use ``np.testing.assert_allclose`` if None. :param ref_fn: the function to generate expected data, should assign output if None. Examples: .. code-block:: dtype = np.float32 cases = [{"input": [10, 20]}, {"input": [20, 30]}] opr_test(cases, F.eye, ref_fn=lambda n, m: np.eye(n, m).astype(dtype), dtype=dtype) """ def check_results(results, expected): if not isinstance(results, (tuple, list)): results = (results,) for r, e in zip(results, expected): if not isinstance(r, (tensor, VarNode)): r = tensor(r) compare_fn(r, e) def get_param(cases, idx): case = cases[idx] inp = case.get("input", None) outp = case.get("output", None) if inp is None: raise ValueError("the test case should have input") if not isinstance(inp, (tuple, list)): inp = (inp,) if ref_fn is not None and callable(ref_fn): outp = ref_fn(*inp) if outp is None: raise ValueError("the test case should have output or reference function") if not isinstance(outp, (tuple, list)): outp = (outp,) return inp, outp if len(cases) == 0: raise ValueError("should give one case at least") if not callable(func): raise ValueError("the input func should be callable") inp, outp = get_param(cases, 0) inp_tensor = [make_tensor(inpi, network) for inpi in inp] if test_trace and not network: copied_inp = inp_tensor.copy() for symbolic in [False, True]: traced_func = trace(symbolic=symbolic)(func) for _ in range(3): traced_results = traced_func(*copied_inp, **kwargs) check_results(traced_results, outp) dumped_func =
trace(symbolic=True, capture_as_const=True)
megengine.jit.trace
import io import numpy as np import megengine.core.tensor.megbrain_graph as G import megengine.utils.comp_graph_tools as cgtools from megengine import tensor from megengine.jit import trace from megengine.utils.network_node import VarNode def _default_compare_fn(x, y): if isinstance(x, np.ndarray): np.testing.assert_allclose(x, y, rtol=1e-6) else: np.testing.assert_allclose(x.numpy(), y, rtol=1e-6) def make_tensor(x, network=None, device=None): if network is not None: if isinstance(x, VarNode): return VarNode(x.var) return network.make_const(x, device=device) else: return tensor(x, device=device) def opr_test( cases, func, compare_fn=_default_compare_fn, ref_fn=None, test_trace=True, network=None, **kwargs ): """ :param cases: the list which have dict element, the list length should be 2 for dynamic shape test. and the dict should have input, and should have output if ref_fn is None. should use list for multiple inputs and outputs for each case. :param func: the function to run opr. :param compare_fn: the function to compare the result and expected, use ``np.testing.assert_allclose`` if None. :param ref_fn: the function to generate expected data, should assign output if None. Examples: .. code-block:: dtype = np.float32 cases = [{"input": [10, 20]}, {"input": [20, 30]}] opr_test(cases, F.eye, ref_fn=lambda n, m: np.eye(n, m).astype(dtype), dtype=dtype) """ def check_results(results, expected): if not isinstance(results, (tuple, list)): results = (results,) for r, e in zip(results, expected): if not isinstance(r, (tensor, VarNode)): r =
tensor(r)
megengine.tensor
import io import numpy as np import megengine.core.tensor.megbrain_graph as G import megengine.utils.comp_graph_tools as cgtools from megengine import tensor from megengine.jit import trace from megengine.utils.network_node import VarNode def _default_compare_fn(x, y): if isinstance(x, np.ndarray): np.testing.assert_allclose(x, y, rtol=1e-6) else: np.testing.assert_allclose(x.numpy(), y, rtol=1e-6) def make_tensor(x, network=None, device=None): if network is not None: if isinstance(x, VarNode): return VarNode(x.var) return network.make_const(x, device=device) else: return tensor(x, device=device) def opr_test( cases, func, compare_fn=_default_compare_fn, ref_fn=None, test_trace=True, network=None, **kwargs ): """ :param cases: the list which have dict element, the list length should be 2 for dynamic shape test. and the dict should have input, and should have output if ref_fn is None. should use list for multiple inputs and outputs for each case. :param func: the function to run opr. :param compare_fn: the function to compare the result and expected, use ``np.testing.assert_allclose`` if None. :param ref_fn: the function to generate expected data, should assign output if None. Examples: .. code-block:: dtype = np.float32 cases = [{"input": [10, 20]}, {"input": [20, 30]}] opr_test(cases, F.eye, ref_fn=lambda n, m: np.eye(n, m).astype(dtype), dtype=dtype) """ def check_results(results, expected): if not isinstance(results, (tuple, list)): results = (results,) for r, e in zip(results, expected): if not isinstance(r, (tensor, VarNode)): r = tensor(r) compare_fn(r, e) def get_param(cases, idx): case = cases[idx] inp = case.get("input", None) outp = case.get("output", None) if inp is None: raise ValueError("the test case should have input") if not isinstance(inp, (tuple, list)): inp = (inp,) if ref_fn is not None and callable(ref_fn): outp = ref_fn(*inp) if outp is None: raise ValueError("the test case should have output or reference function") if not isinstance(outp, (tuple, list)): outp = (outp,) return inp, outp if len(cases) == 0: raise ValueError("should give one case at least") if not callable(func): raise ValueError("the input func should be callable") inp, outp = get_param(cases, 0) inp_tensor = [make_tensor(inpi, network) for inpi in inp] if test_trace and not network: copied_inp = inp_tensor.copy() for symbolic in [False, True]: traced_func =
trace(symbolic=symbolic)
megengine.jit.trace
import numpy as np import megengine import megengine.module as M import megengine.functional as F from edit.models.common import ShuffleV2Block, CoordAtt import math from . import default_init_weights class MobileNeXt(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): """ 默认使用coordinate attention在第一个dwise之后 https://github.com/Andrew-Qibin/CoordAttention/blob/main/coordatt.py """ super(MobileNeXt, self).__init__() self.dconv1 =
M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=in_channels)
megengine.module.ConvRelu2d
import numpy as np import megengine import megengine.module as M import megengine.functional as F from edit.models.common import ShuffleV2Block, CoordAtt import math from . import default_init_weights class MobileNeXt(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): """ 默认使用coordinate attention在第一个dwise之后 https://github.com/Andrew-Qibin/CoordAttention/blob/main/coordatt.py """ super(MobileNeXt, self).__init__() self.dconv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=in_channels) self.CA = CoordAtt(inp = out_channels, oup=out_channels) self.conv1 =
M.Conv2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0)
megengine.module.Conv2d
import numpy as np import megengine import megengine.module as M import megengine.functional as F from edit.models.common import ShuffleV2Block, CoordAtt import math from . import default_init_weights class MobileNeXt(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): """ 默认使用coordinate attention在第一个dwise之后 https://github.com/Andrew-Qibin/CoordAttention/blob/main/coordatt.py """ super(MobileNeXt, self).__init__() self.dconv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=in_channels) self.CA = CoordAtt(inp = out_channels, oup=out_channels) self.conv1 = M.Conv2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.conv2 =
M.ConvRelu2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0)
megengine.module.ConvRelu2d
import numpy as np import megengine import megengine.module as M import megengine.functional as F from edit.models.common import ShuffleV2Block, CoordAtt import math from . import default_init_weights class MobileNeXt(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): """ 默认使用coordinate attention在第一个dwise之后 https://github.com/Andrew-Qibin/CoordAttention/blob/main/coordatt.py """ super(MobileNeXt, self).__init__() self.dconv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=in_channels) self.CA = CoordAtt(inp = out_channels, oup=out_channels) self.conv1 = M.Conv2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.conv2 = M.ConvRelu2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.dconv2 =
M.Conv2d(out_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=out_channels)
megengine.module.Conv2d
import numpy as np import megengine import megengine.module as M import megengine.functional as F from edit.models.common import ShuffleV2Block, CoordAtt import math from . import default_init_weights class MobileNeXt(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): """ 默认使用coordinate attention在第一个dwise之后 https://github.com/Andrew-Qibin/CoordAttention/blob/main/coordatt.py """ super(MobileNeXt, self).__init__() self.dconv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=in_channels) self.CA = CoordAtt(inp = out_channels, oup=out_channels) self.conv1 = M.Conv2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.conv2 = M.ConvRelu2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.dconv2 = M.Conv2d(out_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=out_channels) self.init_weights() def init_weights(self): for m in [self.conv1, self.conv2, self.dconv1, self.dconv2]: default_init_weights(m, scale=0.1) def forward(self, x): identity = x out = self.dconv2(self.conv2(self.conv1(self.CA(self.dconv1(x))))) return identity + out class ResBlock(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): super(ResBlock, self).__init__() self.conv1 =
M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2))
megengine.module.ConvRelu2d
import numpy as np import megengine import megengine.module as M import megengine.functional as F from edit.models.common import ShuffleV2Block, CoordAtt import math from . import default_init_weights class MobileNeXt(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): """ 默认使用coordinate attention在第一个dwise之后 https://github.com/Andrew-Qibin/CoordAttention/blob/main/coordatt.py """ super(MobileNeXt, self).__init__() self.dconv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=in_channels) self.CA = CoordAtt(inp = out_channels, oup=out_channels) self.conv1 = M.Conv2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.conv2 = M.ConvRelu2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.dconv2 = M.Conv2d(out_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=out_channels) self.init_weights() def init_weights(self): for m in [self.conv1, self.conv2, self.dconv1, self.dconv2]: default_init_weights(m, scale=0.1) def forward(self, x): identity = x out = self.dconv2(self.conv2(self.conv1(self.CA(self.dconv1(x))))) return identity + out class ResBlock(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): super(ResBlock, self).__init__() self.conv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2)) self.conv2 =
M.Conv2d(out_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2))
megengine.module.Conv2d
import numpy as np import megengine import megengine.module as M import megengine.functional as F from edit.models.common import ShuffleV2Block, CoordAtt import math from . import default_init_weights class MobileNeXt(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): """ 默认使用coordinate attention在第一个dwise之后 https://github.com/Andrew-Qibin/CoordAttention/blob/main/coordatt.py """ super(MobileNeXt, self).__init__() self.dconv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=in_channels) self.CA = CoordAtt(inp = out_channels, oup=out_channels) self.conv1 = M.Conv2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.conv2 = M.ConvRelu2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.dconv2 = M.Conv2d(out_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=out_channels) self.init_weights() def init_weights(self): for m in [self.conv1, self.conv2, self.dconv1, self.dconv2]: default_init_weights(m, scale=0.1) def forward(self, x): identity = x out = self.dconv2(self.conv2(self.conv1(self.CA(self.dconv1(x))))) return identity + out class ResBlock(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): super(ResBlock, self).__init__() self.conv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2)) self.conv2 = M.Conv2d(out_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2)) self.init_weights() def init_weights(self): for m in [self.conv1, self.conv2]: default_init_weights(m, scale=0.1) def forward(self, x): identity = x out = self.conv2(self.conv1(x)) return identity + out class ResBlocks(M.Module): def __init__(self, channel_num, resblock_num, kernel_size=3, blocktype="resblock"): super(ResBlocks, self).__init__() assert blocktype in ("resblock", "shuffleblock", "MobileNeXt") if blocktype == "resblock": self.model = M.Sequential( self.make_resblock_layer(channel_num, resblock_num, kernel_size), ) elif blocktype == "shuffleblock": self.model = M.Sequential( self.make_shuffleblock_layer(channel_num, resblock_num, kernel_size), ) elif blocktype == "MobileNeXt": self.model = M.Sequential( self.make_MobileNeXt_layer(channel_num, resblock_num, kernel_size) ) else: raise NotImplementedError("") def make_MobileNeXt_layer(self, ch_out, num_blocks, kernel_size): layers = [] for _ in range(num_blocks): layers.append(MobileNeXt(ch_out, ch_out, kernel_size)) return
M.Sequential(*layers)
megengine.module.Sequential
import numpy as np import megengine import megengine.module as M import megengine.functional as F from edit.models.common import ShuffleV2Block, CoordAtt import math from . import default_init_weights class MobileNeXt(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): """ 默认使用coordinate attention在第一个dwise之后 https://github.com/Andrew-Qibin/CoordAttention/blob/main/coordatt.py """ super(MobileNeXt, self).__init__() self.dconv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=in_channels) self.CA = CoordAtt(inp = out_channels, oup=out_channels) self.conv1 = M.Conv2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.conv2 = M.ConvRelu2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.dconv2 = M.Conv2d(out_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=out_channels) self.init_weights() def init_weights(self): for m in [self.conv1, self.conv2, self.dconv1, self.dconv2]: default_init_weights(m, scale=0.1) def forward(self, x): identity = x out = self.dconv2(self.conv2(self.conv1(self.CA(self.dconv1(x))))) return identity + out class ResBlock(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): super(ResBlock, self).__init__() self.conv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2)) self.conv2 = M.Conv2d(out_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2)) self.init_weights() def init_weights(self): for m in [self.conv1, self.conv2]: default_init_weights(m, scale=0.1) def forward(self, x): identity = x out = self.conv2(self.conv1(x)) return identity + out class ResBlocks(M.Module): def __init__(self, channel_num, resblock_num, kernel_size=3, blocktype="resblock"): super(ResBlocks, self).__init__() assert blocktype in ("resblock", "shuffleblock", "MobileNeXt") if blocktype == "resblock": self.model = M.Sequential( self.make_resblock_layer(channel_num, resblock_num, kernel_size), ) elif blocktype == "shuffleblock": self.model = M.Sequential( self.make_shuffleblock_layer(channel_num, resblock_num, kernel_size), ) elif blocktype == "MobileNeXt": self.model = M.Sequential( self.make_MobileNeXt_layer(channel_num, resblock_num, kernel_size) ) else: raise NotImplementedError("") def make_MobileNeXt_layer(self, ch_out, num_blocks, kernel_size): layers = [] for _ in range(num_blocks): layers.append(MobileNeXt(ch_out, ch_out, kernel_size)) return M.Sequential(*layers) def make_resblock_layer(self, ch_out, num_blocks, kernel_size): layers = [] for _ in range(num_blocks): layers.append(ResBlock(ch_out, ch_out, kernel_size)) return
M.Sequential(*layers)
megengine.module.Sequential
import numpy as np import megengine import megengine.module as M import megengine.functional as F from edit.models.common import ShuffleV2Block, CoordAtt import math from . import default_init_weights class MobileNeXt(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): """ 默认使用coordinate attention在第一个dwise之后 https://github.com/Andrew-Qibin/CoordAttention/blob/main/coordatt.py """ super(MobileNeXt, self).__init__() self.dconv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=in_channels) self.CA = CoordAtt(inp = out_channels, oup=out_channels) self.conv1 = M.Conv2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.conv2 = M.ConvRelu2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.dconv2 = M.Conv2d(out_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=out_channels) self.init_weights() def init_weights(self): for m in [self.conv1, self.conv2, self.dconv1, self.dconv2]: default_init_weights(m, scale=0.1) def forward(self, x): identity = x out = self.dconv2(self.conv2(self.conv1(self.CA(self.dconv1(x))))) return identity + out class ResBlock(M.Module): def __init__(self, in_channels, out_channels, kernel_size=3): super(ResBlock, self).__init__() self.conv1 = M.ConvRelu2d(in_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2)) self.conv2 = M.Conv2d(out_channels, out_channels, kernel_size=kernel_size, stride=1, padding=(kernel_size//2)) self.init_weights() def init_weights(self): for m in [self.conv1, self.conv2]: default_init_weights(m, scale=0.1) def forward(self, x): identity = x out = self.conv2(self.conv1(x)) return identity + out class ResBlocks(M.Module): def __init__(self, channel_num, resblock_num, kernel_size=3, blocktype="resblock"): super(ResBlocks, self).__init__() assert blocktype in ("resblock", "shuffleblock", "MobileNeXt") if blocktype == "resblock": self.model = M.Sequential( self.make_resblock_layer(channel_num, resblock_num, kernel_size), ) elif blocktype == "shuffleblock": self.model = M.Sequential( self.make_shuffleblock_layer(channel_num, resblock_num, kernel_size), ) elif blocktype == "MobileNeXt": self.model = M.Sequential( self.make_MobileNeXt_layer(channel_num, resblock_num, kernel_size) ) else: raise NotImplementedError("") def make_MobileNeXt_layer(self, ch_out, num_blocks, kernel_size): layers = [] for _ in range(num_blocks): layers.append(MobileNeXt(ch_out, ch_out, kernel_size)) return M.Sequential(*layers) def make_resblock_layer(self, ch_out, num_blocks, kernel_size): layers = [] for _ in range(num_blocks): layers.append(ResBlock(ch_out, ch_out, kernel_size)) return M.Sequential(*layers) def make_shuffleblock_layer(self, ch_out, num_blocks, kernel_size): layers = [] for _ in range(num_blocks): layers.append(ShuffleV2Block(inp = ch_out//2, oup=ch_out, mid_channels=ch_out//2, ksize=kernel_size, stride=1)) return
M.Sequential(*layers)
megengine.module.Sequential
import megengine as mge import megengine.functional as F from megengine.core import tensor from layers.nms import gpu_nms from config import config from det_opr.bbox_opr import bbox_transform_inv_opr, clip_boxes_opr, \ filter_boxes_opr def find_top_rpn_proposals(is_train, rpn_bbox_offsets_list, rpn_cls_prob_list, all_anchors_list, im_info): prev_nms_top_n = config.train_prev_nms_top_n \ if is_train else config.test_prev_nms_top_n post_nms_top_n = config.train_post_nms_top_n \ if is_train else config.test_post_nms_top_n batch_per_gpu = config.batch_per_gpu if is_train else 1 nms_threshold = config.rpn_nms_threshold box_min_size = config.rpn_min_box_size bbox_normalize_targets = config.rpn_bbox_normalize_targets bbox_normalize_means = config.bbox_normalize_means bbox_normalize_stds = config.bbox_normalize_stds list_size = len(rpn_bbox_offsets_list) return_rois = [] return_probs = [] for bid in range(batch_per_gpu): batch_proposals_list = [] batch_probs_list = [] for l in range(list_size): # get proposals and probs offsets = rpn_bbox_offsets_list[l][bid] \ .dimshuffle(1, 2, 0).reshape(-1, 4) if bbox_normalize_targets: std_opr = tensor(config.bbox_normalize_stds[None, :]) mean_opr = tensor(config.bbox_normalize_means[None, :]) pred_offsets = pred_offsets * std_opr pred_offsets = pred_offsets + mean_opr all_anchors = all_anchors_list[l] proposals = bbox_transform_inv_opr(all_anchors, offsets) if config.anchor_within_border: proposals = clip_boxes_opr(proposals, im_info[bid, :]) probs = rpn_cls_prob_list[l][bid] \ .dimshuffle(1,2,0).reshape(-1, 2) probs = F.softmax(probs)[:, 1] # gather the proposals and probs batch_proposals_list.append(proposals) batch_probs_list.append(probs) batch_proposals =
F.concat(batch_proposals_list, axis=0)
megengine.functional.concat
import megengine as mge import megengine.functional as F from megengine.core import tensor from layers.nms import gpu_nms from config import config from det_opr.bbox_opr import bbox_transform_inv_opr, clip_boxes_opr, \ filter_boxes_opr def find_top_rpn_proposals(is_train, rpn_bbox_offsets_list, rpn_cls_prob_list, all_anchors_list, im_info): prev_nms_top_n = config.train_prev_nms_top_n \ if is_train else config.test_prev_nms_top_n post_nms_top_n = config.train_post_nms_top_n \ if is_train else config.test_post_nms_top_n batch_per_gpu = config.batch_per_gpu if is_train else 1 nms_threshold = config.rpn_nms_threshold box_min_size = config.rpn_min_box_size bbox_normalize_targets = config.rpn_bbox_normalize_targets bbox_normalize_means = config.bbox_normalize_means bbox_normalize_stds = config.bbox_normalize_stds list_size = len(rpn_bbox_offsets_list) return_rois = [] return_probs = [] for bid in range(batch_per_gpu): batch_proposals_list = [] batch_probs_list = [] for l in range(list_size): # get proposals and probs offsets = rpn_bbox_offsets_list[l][bid] \ .dimshuffle(1, 2, 0).reshape(-1, 4) if bbox_normalize_targets: std_opr = tensor(config.bbox_normalize_stds[None, :]) mean_opr = tensor(config.bbox_normalize_means[None, :]) pred_offsets = pred_offsets * std_opr pred_offsets = pred_offsets + mean_opr all_anchors = all_anchors_list[l] proposals = bbox_transform_inv_opr(all_anchors, offsets) if config.anchor_within_border: proposals = clip_boxes_opr(proposals, im_info[bid, :]) probs = rpn_cls_prob_list[l][bid] \ .dimshuffle(1,2,0).reshape(-1, 2) probs = F.softmax(probs)[:, 1] # gather the proposals and probs batch_proposals_list.append(proposals) batch_probs_list.append(probs) batch_proposals = F.concat(batch_proposals_list, axis=0) batch_probs =
F.concat(batch_probs_list, axis=0)
megengine.functional.concat
import megengine as mge import megengine.functional as F from megengine.core import tensor from layers.nms import gpu_nms from config import config from det_opr.bbox_opr import bbox_transform_inv_opr, clip_boxes_opr, \ filter_boxes_opr def find_top_rpn_proposals(is_train, rpn_bbox_offsets_list, rpn_cls_prob_list, all_anchors_list, im_info): prev_nms_top_n = config.train_prev_nms_top_n \ if is_train else config.test_prev_nms_top_n post_nms_top_n = config.train_post_nms_top_n \ if is_train else config.test_post_nms_top_n batch_per_gpu = config.batch_per_gpu if is_train else 1 nms_threshold = config.rpn_nms_threshold box_min_size = config.rpn_min_box_size bbox_normalize_targets = config.rpn_bbox_normalize_targets bbox_normalize_means = config.bbox_normalize_means bbox_normalize_stds = config.bbox_normalize_stds list_size = len(rpn_bbox_offsets_list) return_rois = [] return_probs = [] for bid in range(batch_per_gpu): batch_proposals_list = [] batch_probs_list = [] for l in range(list_size): # get proposals and probs offsets = rpn_bbox_offsets_list[l][bid] \ .dimshuffle(1, 2, 0).reshape(-1, 4) if bbox_normalize_targets: std_opr = tensor(config.bbox_normalize_stds[None, :]) mean_opr = tensor(config.bbox_normalize_means[None, :]) pred_offsets = pred_offsets * std_opr pred_offsets = pred_offsets + mean_opr all_anchors = all_anchors_list[l] proposals = bbox_transform_inv_opr(all_anchors, offsets) if config.anchor_within_border: proposals = clip_boxes_opr(proposals, im_info[bid, :]) probs = rpn_cls_prob_list[l][bid] \ .dimshuffle(1,2,0).reshape(-1, 2) probs = F.softmax(probs)[:, 1] # gather the proposals and probs batch_proposals_list.append(proposals) batch_probs_list.append(probs) batch_proposals = F.concat(batch_proposals_list, axis=0) batch_probs = F.concat(batch_probs_list, axis=0) # filter the zero boxes. batch_keep_mask = filter_boxes_opr( batch_proposals, box_min_size * im_info[bid, 2]) batch_probs = batch_probs * batch_keep_mask # prev_nms_top_n num_proposals = F.minimum(prev_nms_top_n, batch_probs.shapeof()[0]) batch_probs, idx =
F.argsort(batch_probs, descending=True)
megengine.functional.argsort
import megengine as mge import megengine.functional as F from megengine.core import tensor from layers.nms import gpu_nms from config import config from det_opr.bbox_opr import bbox_transform_inv_opr, clip_boxes_opr, \ filter_boxes_opr def find_top_rpn_proposals(is_train, rpn_bbox_offsets_list, rpn_cls_prob_list, all_anchors_list, im_info): prev_nms_top_n = config.train_prev_nms_top_n \ if is_train else config.test_prev_nms_top_n post_nms_top_n = config.train_post_nms_top_n \ if is_train else config.test_post_nms_top_n batch_per_gpu = config.batch_per_gpu if is_train else 1 nms_threshold = config.rpn_nms_threshold box_min_size = config.rpn_min_box_size bbox_normalize_targets = config.rpn_bbox_normalize_targets bbox_normalize_means = config.bbox_normalize_means bbox_normalize_stds = config.bbox_normalize_stds list_size = len(rpn_bbox_offsets_list) return_rois = [] return_probs = [] for bid in range(batch_per_gpu): batch_proposals_list = [] batch_probs_list = [] for l in range(list_size): # get proposals and probs offsets = rpn_bbox_offsets_list[l][bid] \ .dimshuffle(1, 2, 0).reshape(-1, 4) if bbox_normalize_targets: std_opr = tensor(config.bbox_normalize_stds[None, :]) mean_opr = tensor(config.bbox_normalize_means[None, :]) pred_offsets = pred_offsets * std_opr pred_offsets = pred_offsets + mean_opr all_anchors = all_anchors_list[l] proposals = bbox_transform_inv_opr(all_anchors, offsets) if config.anchor_within_border: proposals = clip_boxes_opr(proposals, im_info[bid, :]) probs = rpn_cls_prob_list[l][bid] \ .dimshuffle(1,2,0).reshape(-1, 2) probs = F.softmax(probs)[:, 1] # gather the proposals and probs batch_proposals_list.append(proposals) batch_probs_list.append(probs) batch_proposals = F.concat(batch_proposals_list, axis=0) batch_probs = F.concat(batch_probs_list, axis=0) # filter the zero boxes. batch_keep_mask = filter_boxes_opr( batch_proposals, box_min_size * im_info[bid, 2]) batch_probs = batch_probs * batch_keep_mask # prev_nms_top_n num_proposals = F.minimum(prev_nms_top_n, batch_probs.shapeof()[0]) batch_probs, idx = F.argsort(batch_probs, descending=True) batch_probs = batch_probs[:num_proposals].reshape(-1,1) topk_idx = idx[:num_proposals].reshape(-1) batch_proposals = batch_proposals.ai[topk_idx] batch_rois =
F.concat([batch_proposals, batch_probs], axis=1)
megengine.functional.concat
import megengine as mge import megengine.functional as F from megengine.core import tensor from layers.nms import gpu_nms from config import config from det_opr.bbox_opr import bbox_transform_inv_opr, clip_boxes_opr, \ filter_boxes_opr def find_top_rpn_proposals(is_train, rpn_bbox_offsets_list, rpn_cls_prob_list, all_anchors_list, im_info): prev_nms_top_n = config.train_prev_nms_top_n \ if is_train else config.test_prev_nms_top_n post_nms_top_n = config.train_post_nms_top_n \ if is_train else config.test_post_nms_top_n batch_per_gpu = config.batch_per_gpu if is_train else 1 nms_threshold = config.rpn_nms_threshold box_min_size = config.rpn_min_box_size bbox_normalize_targets = config.rpn_bbox_normalize_targets bbox_normalize_means = config.bbox_normalize_means bbox_normalize_stds = config.bbox_normalize_stds list_size = len(rpn_bbox_offsets_list) return_rois = [] return_probs = [] for bid in range(batch_per_gpu): batch_proposals_list = [] batch_probs_list = [] for l in range(list_size): # get proposals and probs offsets = rpn_bbox_offsets_list[l][bid] \ .dimshuffle(1, 2, 0).reshape(-1, 4) if bbox_normalize_targets: std_opr = tensor(config.bbox_normalize_stds[None, :]) mean_opr = tensor(config.bbox_normalize_means[None, :]) pred_offsets = pred_offsets * std_opr pred_offsets = pred_offsets + mean_opr all_anchors = all_anchors_list[l] proposals = bbox_transform_inv_opr(all_anchors, offsets) if config.anchor_within_border: proposals = clip_boxes_opr(proposals, im_info[bid, :]) probs = rpn_cls_prob_list[l][bid] \ .dimshuffle(1,2,0).reshape(-1, 2) probs = F.softmax(probs)[:, 1] # gather the proposals and probs batch_proposals_list.append(proposals) batch_probs_list.append(probs) batch_proposals = F.concat(batch_proposals_list, axis=0) batch_probs = F.concat(batch_probs_list, axis=0) # filter the zero boxes. batch_keep_mask = filter_boxes_opr( batch_proposals, box_min_size * im_info[bid, 2]) batch_probs = batch_probs * batch_keep_mask # prev_nms_top_n num_proposals = F.minimum(prev_nms_top_n, batch_probs.shapeof()[0]) batch_probs, idx = F.argsort(batch_probs, descending=True) batch_probs = batch_probs[:num_proposals].reshape(-1,1) topk_idx = idx[:num_proposals].reshape(-1) batch_proposals = batch_proposals.ai[topk_idx] batch_rois = F.concat([batch_proposals, batch_probs], axis=1) # For each image, run a total-level NMS, and choose topk results. keep_inds = gpu_nms(batch_rois, nms_threshold, post_nms_top_n) batch_rois = batch_rois.ai[keep_inds] batch_probs = batch_rois[:, -1] # cons the rois batch_inds = mge.ones((batch_rois.shapeof()[0], 1)) * bid batch_rois =
F.concat([batch_inds, batch_rois[:, :-1]], axis=1)
megengine.functional.concat
import megengine as mge import megengine.functional as F from megengine.core import tensor from layers.nms import gpu_nms from config import config from det_opr.bbox_opr import bbox_transform_inv_opr, clip_boxes_opr, \ filter_boxes_opr def find_top_rpn_proposals(is_train, rpn_bbox_offsets_list, rpn_cls_prob_list, all_anchors_list, im_info): prev_nms_top_n = config.train_prev_nms_top_n \ if is_train else config.test_prev_nms_top_n post_nms_top_n = config.train_post_nms_top_n \ if is_train else config.test_post_nms_top_n batch_per_gpu = config.batch_per_gpu if is_train else 1 nms_threshold = config.rpn_nms_threshold box_min_size = config.rpn_min_box_size bbox_normalize_targets = config.rpn_bbox_normalize_targets bbox_normalize_means = config.bbox_normalize_means bbox_normalize_stds = config.bbox_normalize_stds list_size = len(rpn_bbox_offsets_list) return_rois = [] return_probs = [] for bid in range(batch_per_gpu): batch_proposals_list = [] batch_probs_list = [] for l in range(list_size): # get proposals and probs offsets = rpn_bbox_offsets_list[l][bid] \ .dimshuffle(1, 2, 0).reshape(-1, 4) if bbox_normalize_targets: std_opr = tensor(config.bbox_normalize_stds[None, :]) mean_opr = tensor(config.bbox_normalize_means[None, :]) pred_offsets = pred_offsets * std_opr pred_offsets = pred_offsets + mean_opr all_anchors = all_anchors_list[l] proposals = bbox_transform_inv_opr(all_anchors, offsets) if config.anchor_within_border: proposals = clip_boxes_opr(proposals, im_info[bid, :]) probs = rpn_cls_prob_list[l][bid] \ .dimshuffle(1,2,0).reshape(-1, 2) probs = F.softmax(probs)[:, 1] # gather the proposals and probs batch_proposals_list.append(proposals) batch_probs_list.append(probs) batch_proposals = F.concat(batch_proposals_list, axis=0) batch_probs = F.concat(batch_probs_list, axis=0) # filter the zero boxes. batch_keep_mask = filter_boxes_opr( batch_proposals, box_min_size * im_info[bid, 2]) batch_probs = batch_probs * batch_keep_mask # prev_nms_top_n num_proposals = F.minimum(prev_nms_top_n, batch_probs.shapeof()[0]) batch_probs, idx = F.argsort(batch_probs, descending=True) batch_probs = batch_probs[:num_proposals].reshape(-1,1) topk_idx = idx[:num_proposals].reshape(-1) batch_proposals = batch_proposals.ai[topk_idx] batch_rois = F.concat([batch_proposals, batch_probs], axis=1) # For each image, run a total-level NMS, and choose topk results. keep_inds = gpu_nms(batch_rois, nms_threshold, post_nms_top_n) batch_rois = batch_rois.ai[keep_inds] batch_probs = batch_rois[:, -1] # cons the rois batch_inds = mge.ones((batch_rois.shapeof()[0], 1)) * bid batch_rois = F.concat([batch_inds, batch_rois[:, :-1]], axis=1) return_rois.append(batch_rois) return_probs.append(batch_probs) if batch_per_gpu == 1: return batch_rois, batch_probs else: concated_rois =
F.concat(return_rois, axis=0)
megengine.functional.concat
import megengine as mge import megengine.functional as F from megengine.core import tensor from layers.nms import gpu_nms from config import config from det_opr.bbox_opr import bbox_transform_inv_opr, clip_boxes_opr, \ filter_boxes_opr def find_top_rpn_proposals(is_train, rpn_bbox_offsets_list, rpn_cls_prob_list, all_anchors_list, im_info): prev_nms_top_n = config.train_prev_nms_top_n \ if is_train else config.test_prev_nms_top_n post_nms_top_n = config.train_post_nms_top_n \ if is_train else config.test_post_nms_top_n batch_per_gpu = config.batch_per_gpu if is_train else 1 nms_threshold = config.rpn_nms_threshold box_min_size = config.rpn_min_box_size bbox_normalize_targets = config.rpn_bbox_normalize_targets bbox_normalize_means = config.bbox_normalize_means bbox_normalize_stds = config.bbox_normalize_stds list_size = len(rpn_bbox_offsets_list) return_rois = [] return_probs = [] for bid in range(batch_per_gpu): batch_proposals_list = [] batch_probs_list = [] for l in range(list_size): # get proposals and probs offsets = rpn_bbox_offsets_list[l][bid] \ .dimshuffle(1, 2, 0).reshape(-1, 4) if bbox_normalize_targets: std_opr = tensor(config.bbox_normalize_stds[None, :]) mean_opr = tensor(config.bbox_normalize_means[None, :]) pred_offsets = pred_offsets * std_opr pred_offsets = pred_offsets + mean_opr all_anchors = all_anchors_list[l] proposals = bbox_transform_inv_opr(all_anchors, offsets) if config.anchor_within_border: proposals = clip_boxes_opr(proposals, im_info[bid, :]) probs = rpn_cls_prob_list[l][bid] \ .dimshuffle(1,2,0).reshape(-1, 2) probs = F.softmax(probs)[:, 1] # gather the proposals and probs batch_proposals_list.append(proposals) batch_probs_list.append(probs) batch_proposals = F.concat(batch_proposals_list, axis=0) batch_probs = F.concat(batch_probs_list, axis=0) # filter the zero boxes. batch_keep_mask = filter_boxes_opr( batch_proposals, box_min_size * im_info[bid, 2]) batch_probs = batch_probs * batch_keep_mask # prev_nms_top_n num_proposals = F.minimum(prev_nms_top_n, batch_probs.shapeof()[0]) batch_probs, idx = F.argsort(batch_probs, descending=True) batch_probs = batch_probs[:num_proposals].reshape(-1,1) topk_idx = idx[:num_proposals].reshape(-1) batch_proposals = batch_proposals.ai[topk_idx] batch_rois = F.concat([batch_proposals, batch_probs], axis=1) # For each image, run a total-level NMS, and choose topk results. keep_inds = gpu_nms(batch_rois, nms_threshold, post_nms_top_n) batch_rois = batch_rois.ai[keep_inds] batch_probs = batch_rois[:, -1] # cons the rois batch_inds = mge.ones((batch_rois.shapeof()[0], 1)) * bid batch_rois = F.concat([batch_inds, batch_rois[:, :-1]], axis=1) return_rois.append(batch_rois) return_probs.append(batch_probs) if batch_per_gpu == 1: return batch_rois, batch_probs else: concated_rois = F.concat(return_rois, axis=0) concated_probs =
F.concat(return_probs, axis=0)
megengine.functional.concat
import megengine as mge import megengine.functional as F from megengine.core import tensor from layers.nms import gpu_nms from config import config from det_opr.bbox_opr import bbox_transform_inv_opr, clip_boxes_opr, \ filter_boxes_opr def find_top_rpn_proposals(is_train, rpn_bbox_offsets_list, rpn_cls_prob_list, all_anchors_list, im_info): prev_nms_top_n = config.train_prev_nms_top_n \ if is_train else config.test_prev_nms_top_n post_nms_top_n = config.train_post_nms_top_n \ if is_train else config.test_post_nms_top_n batch_per_gpu = config.batch_per_gpu if is_train else 1 nms_threshold = config.rpn_nms_threshold box_min_size = config.rpn_min_box_size bbox_normalize_targets = config.rpn_bbox_normalize_targets bbox_normalize_means = config.bbox_normalize_means bbox_normalize_stds = config.bbox_normalize_stds list_size = len(rpn_bbox_offsets_list) return_rois = [] return_probs = [] for bid in range(batch_per_gpu): batch_proposals_list = [] batch_probs_list = [] for l in range(list_size): # get proposals and probs offsets = rpn_bbox_offsets_list[l][bid] \ .dimshuffle(1, 2, 0).reshape(-1, 4) if bbox_normalize_targets: std_opr =
tensor(config.bbox_normalize_stds[None, :])
megengine.core.tensor
import megengine as mge import megengine.functional as F from megengine.core import tensor from layers.nms import gpu_nms from config import config from det_opr.bbox_opr import bbox_transform_inv_opr, clip_boxes_opr, \ filter_boxes_opr def find_top_rpn_proposals(is_train, rpn_bbox_offsets_list, rpn_cls_prob_list, all_anchors_list, im_info): prev_nms_top_n = config.train_prev_nms_top_n \ if is_train else config.test_prev_nms_top_n post_nms_top_n = config.train_post_nms_top_n \ if is_train else config.test_post_nms_top_n batch_per_gpu = config.batch_per_gpu if is_train else 1 nms_threshold = config.rpn_nms_threshold box_min_size = config.rpn_min_box_size bbox_normalize_targets = config.rpn_bbox_normalize_targets bbox_normalize_means = config.bbox_normalize_means bbox_normalize_stds = config.bbox_normalize_stds list_size = len(rpn_bbox_offsets_list) return_rois = [] return_probs = [] for bid in range(batch_per_gpu): batch_proposals_list = [] batch_probs_list = [] for l in range(list_size): # get proposals and probs offsets = rpn_bbox_offsets_list[l][bid] \ .dimshuffle(1, 2, 0).reshape(-1, 4) if bbox_normalize_targets: std_opr = tensor(config.bbox_normalize_stds[None, :]) mean_opr =
tensor(config.bbox_normalize_means[None, :])
megengine.core.tensor
import megengine as mge import megengine.functional as F from megengine.core import tensor from layers.nms import gpu_nms from config import config from det_opr.bbox_opr import bbox_transform_inv_opr, clip_boxes_opr, \ filter_boxes_opr def find_top_rpn_proposals(is_train, rpn_bbox_offsets_list, rpn_cls_prob_list, all_anchors_list, im_info): prev_nms_top_n = config.train_prev_nms_top_n \ if is_train else config.test_prev_nms_top_n post_nms_top_n = config.train_post_nms_top_n \ if is_train else config.test_post_nms_top_n batch_per_gpu = config.batch_per_gpu if is_train else 1 nms_threshold = config.rpn_nms_threshold box_min_size = config.rpn_min_box_size bbox_normalize_targets = config.rpn_bbox_normalize_targets bbox_normalize_means = config.bbox_normalize_means bbox_normalize_stds = config.bbox_normalize_stds list_size = len(rpn_bbox_offsets_list) return_rois = [] return_probs = [] for bid in range(batch_per_gpu): batch_proposals_list = [] batch_probs_list = [] for l in range(list_size): # get proposals and probs offsets = rpn_bbox_offsets_list[l][bid] \ .dimshuffle(1, 2, 0).reshape(-1, 4) if bbox_normalize_targets: std_opr = tensor(config.bbox_normalize_stds[None, :]) mean_opr = tensor(config.bbox_normalize_means[None, :]) pred_offsets = pred_offsets * std_opr pred_offsets = pred_offsets + mean_opr all_anchors = all_anchors_list[l] proposals = bbox_transform_inv_opr(all_anchors, offsets) if config.anchor_within_border: proposals = clip_boxes_opr(proposals, im_info[bid, :]) probs = rpn_cls_prob_list[l][bid] \ .dimshuffle(1,2,0).reshape(-1, 2) probs =
F.softmax(probs)
megengine.functional.softmax
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre =
F.stack([ptrx, ptry], axis=1)
megengine.functional.stack
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes =
F.concat([gt_boxes, dummy], axis=0)
megengine.functional.concat
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers =
F.expand_dims(anchor_centers, axis=1)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers =
F.expand_dims(gtboxes_centers, axis=0)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance =
F.abs(an_centers - gt_centers)
megengine.functional.abs
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *=
F.expand_dims(valid_mask, axis=0)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious =
F.concat(ious_list, axis=0)
megengine.functional.concat
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var =
F.std(ious, 0)
megengine.functional.std
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt =
F.maximum(iou_thresh_per_gt, 0.2)
megengine.functional.maximum
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt =
F.stack([l, r, t, b], axis=2)
megengine.functional.stack
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index =
F.argsort(ious, 1)
megengine.functional.argsort
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps =
F.gather(ious, 1, sorted_index)
megengine.functional.gather
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers =
F.expand_dims(anchor_centers, axis=1)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers =
F.expand_dims(gtboxes_centers, axis=0)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance =
F.abs(an_centers - gt_centers)
megengine.functional.abs
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious =
F.concat(ious_list, axis=0)
megengine.functional.concat
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt =
F.stack([l, r, t, b], axis=2)
megengine.functional.stack
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(F.expand_dims(rpn_target_boxes, 0)) rpn_labels = F.concat(rpn_label_list, axis = 0) rpn_target_boxes = F.concat(rpn_target_boxes_list, axis = 0) return rpn_labels, rpn_target_boxes def mask_anchor_opr(gtboxes, im_info, anchors, labels): eps = 1e-6 gtboxes = gtboxes[:im_info[5].astype(np.int32), :] ignore_mask = (gtboxes[:, 4] < 0).astype(np.float32) mask_flag =
F.zeros(labels.shape[0])
megengine.functional.zeros
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers *
F.expand_dims(valid_mask, axis=1)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index =
F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4])
megengine.functional.cond_take
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index =
F.argsort(level_dist, descending=False)
megengine.functional.argsort
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(
F.expand_dims(pos_area, axis=0)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) *
F.expand_dims(valid_mask, axis=0)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -
F.ones(n)
megengine.functional.ones
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(F.expand_dims(rpn_target_boxes, 0)) rpn_labels = F.concat(rpn_label_list, axis = 0) rpn_target_boxes = F.concat(rpn_target_boxes_list, axis = 0) return rpn_labels, rpn_target_boxes def mask_anchor_opr(gtboxes, im_info, anchors, labels): eps = 1e-6 gtboxes = gtboxes[:im_info[5].astype(np.int32), :] ignore_mask = (gtboxes[:, 4] < 0).astype(np.float32) mask_flag = F.zeros(labels.shape[0]) N, K = anchors.shape[0], gtboxes.shape[0] p_pred = F.broadcast_to(
F.expand_dims(anchors, 1)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(F.expand_dims(rpn_target_boxes, 0)) rpn_labels = F.concat(rpn_label_list, axis = 0) rpn_target_boxes = F.concat(rpn_target_boxes_list, axis = 0) return rpn_labels, rpn_target_boxes def mask_anchor_opr(gtboxes, im_info, anchors, labels): eps = 1e-6 gtboxes = gtboxes[:im_info[5].astype(np.int32), :] ignore_mask = (gtboxes[:, 4] < 0).astype(np.float32) mask_flag = F.zeros(labels.shape[0]) N, K = anchors.shape[0], gtboxes.shape[0] p_pred = F.broadcast_to(F.expand_dims(anchors, 1), (N, K, anchors.shape[1])) p_gt = F.broadcast_to(
F.expand_dims(gtboxes, 0)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(F.expand_dims(rpn_target_boxes, 0)) rpn_labels = F.concat(rpn_label_list, axis = 0) rpn_target_boxes = F.concat(rpn_target_boxes_list, axis = 0) return rpn_labels, rpn_target_boxes def mask_anchor_opr(gtboxes, im_info, anchors, labels): eps = 1e-6 gtboxes = gtboxes[:im_info[5].astype(np.int32), :] ignore_mask = (gtboxes[:, 4] < 0).astype(np.float32) mask_flag = F.zeros(labels.shape[0]) N, K = anchors.shape[0], gtboxes.shape[0] p_pred = F.broadcast_to(F.expand_dims(anchors, 1), (N, K, anchors.shape[1])) p_gt = F.broadcast_to(F.expand_dims(gtboxes, 0), (N, K, gtboxes.shape[1])) max_off = F.concat([F.maximum(p_pred[:,:, :2], p_gt[:,:,:2]), F.minimum(p_pred[:, :, 2:4], p_gt[:, :, 2:4])], axis = 2) I =
F.maximum(max_off[:, :, 2] - max_off[:, :, 0] + 1, 0)
megengine.functional.maximum
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(F.expand_dims(rpn_target_boxes, 0)) rpn_labels = F.concat(rpn_label_list, axis = 0) rpn_target_boxes = F.concat(rpn_target_boxes_list, axis = 0) return rpn_labels, rpn_target_boxes def mask_anchor_opr(gtboxes, im_info, anchors, labels): eps = 1e-6 gtboxes = gtboxes[:im_info[5].astype(np.int32), :] ignore_mask = (gtboxes[:, 4] < 0).astype(np.float32) mask_flag = F.zeros(labels.shape[0]) N, K = anchors.shape[0], gtboxes.shape[0] p_pred = F.broadcast_to(F.expand_dims(anchors, 1), (N, K, anchors.shape[1])) p_gt = F.broadcast_to(F.expand_dims(gtboxes, 0), (N, K, gtboxes.shape[1])) max_off = F.concat([F.maximum(p_pred[:,:, :2], p_gt[:,:,:2]), F.minimum(p_pred[:, :, 2:4], p_gt[:, :, 2:4])], axis = 2) I = F.maximum(max_off[:, :, 2] - max_off[:, :, 0] + 1, 0) * F.maximum( max_off[:, :, 3] - max_off[:, :, 1] + 1, 0) A =
F.maximum(p_pred[:, :, 2] - p_pred[:, :, 0] + 1, 0)
megengine.functional.maximum
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(F.expand_dims(rpn_target_boxes, 0)) rpn_labels = F.concat(rpn_label_list, axis = 0) rpn_target_boxes = F.concat(rpn_target_boxes_list, axis = 0) return rpn_labels, rpn_target_boxes def mask_anchor_opr(gtboxes, im_info, anchors, labels): eps = 1e-6 gtboxes = gtboxes[:im_info[5].astype(np.int32), :] ignore_mask = (gtboxes[:, 4] < 0).astype(np.float32) mask_flag = F.zeros(labels.shape[0]) N, K = anchors.shape[0], gtboxes.shape[0] p_pred = F.broadcast_to(F.expand_dims(anchors, 1), (N, K, anchors.shape[1])) p_gt = F.broadcast_to(F.expand_dims(gtboxes, 0), (N, K, gtboxes.shape[1])) max_off = F.concat([F.maximum(p_pred[:,:, :2], p_gt[:,:,:2]), F.minimum(p_pred[:, :, 2:4], p_gt[:, :, 2:4])], axis = 2) I = F.maximum(max_off[:, :, 2] - max_off[:, :, 0] + 1, 0) * F.maximum( max_off[:, :, 3] - max_off[:, :, 1] + 1, 0) A = F.maximum(p_pred[:, :, 2] - p_pred[:, :, 0] + 1, 0) * F.maximum( p_pred[:, :, 3] - p_pred[:, :, 1] + 1, 0) # I = F.maximum(I, 0) # A = F.maximum(A, 0) IoA = I / (A + eps) IoA = IoA *
F.expand_dims(ignore_mask, 0)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(F.expand_dims(rpn_target_boxes, 0)) rpn_labels = F.concat(rpn_label_list, axis = 0) rpn_target_boxes = F.concat(rpn_target_boxes_list, axis = 0) return rpn_labels, rpn_target_boxes def mask_anchor_opr(gtboxes, im_info, anchors, labels): eps = 1e-6 gtboxes = gtboxes[:im_info[5].astype(np.int32), :] ignore_mask = (gtboxes[:, 4] < 0).astype(np.float32) mask_flag = F.zeros(labels.shape[0]) N, K = anchors.shape[0], gtboxes.shape[0] p_pred = F.broadcast_to(F.expand_dims(anchors, 1), (N, K, anchors.shape[1])) p_gt = F.broadcast_to(F.expand_dims(gtboxes, 0), (N, K, gtboxes.shape[1])) max_off = F.concat([F.maximum(p_pred[:,:, :2], p_gt[:,:,:2]), F.minimum(p_pred[:, :, 2:4], p_gt[:, :, 2:4])], axis = 2) I = F.maximum(max_off[:, :, 2] - max_off[:, :, 0] + 1, 0) * F.maximum( max_off[:, :, 3] - max_off[:, :, 1] + 1, 0) A = F.maximum(p_pred[:, :, 2] - p_pred[:, :, 0] + 1, 0) * F.maximum( p_pred[:, :, 3] - p_pred[:, :, 1] + 1, 0) # I = F.maximum(I, 0) # A = F.maximum(A, 0) IoA = I / (A + eps) IoA = IoA * F.expand_dims(ignore_mask, 0) mask_flag = (IoA > 0.5).sum(axis=1) > 0 labels = labels - F.equal(labels, 0).astype(np.float32) * mask_flag.astype(np.float32) return labels def rpn_anchor_target_opr_impl( gt_boxes, im_info, anchors, clobber_positives = True, ignore_label=-1, background_label=0): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() anchors = anchors.detach() # NOTE: For multi-gpu version, this function should be re-written a_shp0 = anchors.shape[0] valid_gt_boxes = gt_boxes[:im_info[5], :] valid_mask = (gt_boxes[:im_info[5], 4] > 0).astype(np.float32) overlaps = box_overlap_opr(anchors[:, :4], valid_gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) argmax_overlaps = torch.argmax(overlaps,axis=1) max_overlaps = torch.gather(overlaps, 1, argmax_overlaps.unsqueeze(1)) gt_argmax_overlaps = torch.argmax(overlaps, axis=0) gt_argmax_overlaps = torch.gather(overlaps, 1, gt_argmax_overlaps.unsqueeze(0)) cond_max_overlaps = overlaps.eq(gt_argmax_overlaps).astype(np.float32) cmo_shape1 = cond_max_overlaps.shape[1] gt_argmax_overlaps = torch.nonzero(cond_max_overlaps.flatten(), as_tuple=False) gt_argmax_overlaps = gt_argmax_overlaps // cmo_shape1 labels = ignore_label *
F.ones(a_shp0)
megengine.functional.ones
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >=
F.expand_dims(iou_thresh_per_gt, axis=0)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(
F.expand_dims(rpn_labels, 0)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(
F.expand_dims(rpn_target_boxes, 0)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(F.expand_dims(rpn_target_boxes, 0)) rpn_labels = F.concat(rpn_label_list, axis = 0) rpn_target_boxes = F.concat(rpn_target_boxes_list, axis = 0) return rpn_labels, rpn_target_boxes def mask_anchor_opr(gtboxes, im_info, anchors, labels): eps = 1e-6 gtboxes = gtboxes[:im_info[5].astype(np.int32), :] ignore_mask = (gtboxes[:, 4] < 0).astype(np.float32) mask_flag = F.zeros(labels.shape[0]) N, K = anchors.shape[0], gtboxes.shape[0] p_pred = F.broadcast_to(F.expand_dims(anchors, 1), (N, K, anchors.shape[1])) p_gt = F.broadcast_to(F.expand_dims(gtboxes, 0), (N, K, gtboxes.shape[1])) max_off = F.concat([
F.maximum(p_pred[:,:, :2], p_gt[:,:,:2])
megengine.functional.maximum
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(F.expand_dims(rpn_target_boxes, 0)) rpn_labels = F.concat(rpn_label_list, axis = 0) rpn_target_boxes = F.concat(rpn_target_boxes_list, axis = 0) return rpn_labels, rpn_target_boxes def mask_anchor_opr(gtboxes, im_info, anchors, labels): eps = 1e-6 gtboxes = gtboxes[:im_info[5].astype(np.int32), :] ignore_mask = (gtboxes[:, 4] < 0).astype(np.float32) mask_flag = F.zeros(labels.shape[0]) N, K = anchors.shape[0], gtboxes.shape[0] p_pred = F.broadcast_to(F.expand_dims(anchors, 1), (N, K, anchors.shape[1])) p_gt = F.broadcast_to(F.expand_dims(gtboxes, 0), (N, K, gtboxes.shape[1])) max_off = F.concat([F.maximum(p_pred[:,:, :2], p_gt[:,:,:2]),
F.minimum(p_pred[:, :, 2:4], p_gt[:, :, 2:4])
megengine.functional.minimum
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -
F.ones([1, gt_boxes.shape[1]])
megengine.functional.ones
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(
F.pow(distance, 2)
megengine.functional.pow
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious =
F.gather(ious, 1, sorted_index[:, :n])
megengine.functional.gather
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -
F.ones(2 * n)
megengine.functional.ones
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(
F.expand_dims(all_anchors, axis=1)
megengine.functional.expand_dims
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) -
F.equal(labels, -1)
megengine.functional.equal
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(
F.pow(distance, 2)
megengine.functional.pow
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) * F.expand_dims(ignore_label < 0, 1).astype(np.float32) # rpn_labels = rpn_labels - rpn_labels.eq(0).astype(np.float32) * (ignore_label < 0).unsqueeze(1).astype(np.float32) rpn_label_list.append(F.expand_dims(rpn_labels, 0)) rpn_target_boxes_list.append(F.expand_dims(rpn_target_boxes, 0)) rpn_labels = F.concat(rpn_label_list, axis = 0) rpn_target_boxes = F.concat(rpn_target_boxes_list, axis = 0) return rpn_labels, rpn_target_boxes def mask_anchor_opr(gtboxes, im_info, anchors, labels): eps = 1e-6 gtboxes = gtboxes[:im_info[5].astype(np.int32), :] ignore_mask = (gtboxes[:, 4] < 0).astype(np.float32) mask_flag = F.zeros(labels.shape[0]) N, K = anchors.shape[0], gtboxes.shape[0] p_pred = F.broadcast_to(F.expand_dims(anchors, 1), (N, K, anchors.shape[1])) p_gt = F.broadcast_to(F.expand_dims(gtboxes, 0), (N, K, gtboxes.shape[1])) max_off = F.concat([F.maximum(p_pred[:,:, :2], p_gt[:,:,:2]), F.minimum(p_pred[:, :, 2:4], p_gt[:, :, 2:4])], axis = 2) I = F.maximum(max_off[:, :, 2] - max_off[:, :, 0] + 1, 0) * F.maximum( max_off[:, :, 3] - max_off[:, :, 1] + 1, 0) A = F.maximum(p_pred[:, :, 2] - p_pred[:, :, 0] + 1, 0) * F.maximum( p_pred[:, :, 3] - p_pred[:, :, 1] + 1, 0) # I = F.maximum(I, 0) # A = F.maximum(A, 0) IoA = I / (A + eps) IoA = IoA * F.expand_dims(ignore_mask, 0) mask_flag = (IoA > 0.5).sum(axis=1) > 0 labels = labels -
F.equal(labels, 0)
megengine.functional.equal
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 -
F.equal(labels, -1)
megengine.functional.equal
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels -
F.equal(rpn_labels, 0)
megengine.functional.equal
import os, sys import numpy as np from config import config from det_opr.bbox_opr import box_overlap_opr, bbox_transform_opr import megengine as mge from megengine import functional as F import pdb def _compute_center(boxes): ptrx = 0.5 * (boxes[:, 0] + boxes[:, 2]) ptry = 0.5 * (boxes[:, 1] + boxes[:, 3]) centre = F.stack([ptrx, ptry], axis=1) return centre def _compute_pos_area(gtboxes, ratio = 0.3): H, W = gtboxes[:, 3] - gtboxes[:, 1], gtboxes[:, 2] - gtboxes[:, 0] centres = _compute_center(gtboxes) l = centres[:, 0] - ratio * W r = centres[:, 0] + ratio * W t = centres[:, 1] - ratio * H b = centres[:, 1] + ratio * H boundary = F.stack([l, t, r, b], axis = 1) return boundary def _anchor_double_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5].astype(np.int32), :] dummy = -F.ones([1, gt_boxes.shape[1]]).to(gt_boxes.device) gt_boxes = F.concat([gt_boxes, dummy], axis=0) valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) # gtboxes_centers = gtboxes_centers * valid_mask.unsqueeze(1) gtboxes_centers = gtboxes_centers * F.expand_dims(valid_mask, axis=1) N, K = all_anchors.shape[0], gt_boxes.shape[0] an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps *= F.expand_dims(valid_mask, axis=0) default_num = 16 ious_list = [] for l in range(start, end): _, index = F.cond_take(all_anchors[:, 4] == l, all_anchors[:, 4]) level_dist = distance[index, :].transpose(1, 0) ious = overlaps[index, :].transpose(1, 0) sorted_index = F.argsort(level_dist, descending=False) n = min(sorted_index.shape[1], default_num) ious = F.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = F.mean(ious, axis = 0) std_var = F.std(ious, 0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = F.maximum(iou_thresh_per_gt, 0.2) # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers pos_area = _compute_pos_area(gt_boxes, 0.3) # pos_area = pos_area.unsqueeze(0).repeat(N, 1, 1) pos_area = F.broadcast_to(F.expand_dims(pos_area, axis=0), (N, K, pos_area.shape[-1])) l = anchor_points[:, :, 0] - pos_area[:, :, 0] r = pos_area[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - pos_area[:, :, 1] b = pos_area[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= F.expand_dims(iou_thresh_per_gt, axis=0)) * is_in_gt.astype(np.float32) ious = overlaps * valid_mask sorted_index = F.argsort(ious, 1) sorted_overlaps = F.gather(ious, 1, sorted_index) max_overlaps = sorted_overlaps[:, :2].flatten() argmax_overlaps = sorted_index[:, :2].flatten() n, c = all_anchors.shape device = all_anchors.device labels = -F.ones(2 * n).to(device) positive_mask = (max_overlaps >= 0.2).to(device).astype(np.float32) negative_mask = (max_overlaps < 0.2).to(device).astype(np.float32) labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] all_anchors = F.broadcast_to(F.expand_dims(all_anchors, axis=1), (n,2, c)).reshape(-1, c) bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - F.equal(labels, -1).astype(np.float32)) - F.equal(labels, -1).astype(np.float32) return labels, bbox_targets, labels_cat def _anchor_target(gt_boxes, im_info, all_anchors): gt_boxes, im_info = gt_boxes.detach(), im_info.detach() all_anchors = all_anchors.detach() gt_boxes = gt_boxes[:im_info[5], :] valid_mask = 1 - (gt_boxes[:, 4] < 0).astype(np.float32) anchor_centers = _compute_center(all_anchors) gtboxes_centers = _compute_center(gt_boxes) * F.expand_dims(valid_mask, axis=0) N, K = all_anchors.shape[0], gt_boxes.shape[0] # an_centers = anchor_centers.unsqueeze(1).repeat(1, K, 1) an_centers = F.expand_dims(anchor_centers, axis=1) gt_centers = F.expand_dims(gtboxes_centers, axis=0) # gt_centers = gtboxes_centers.unsqueeze(0).repeat(N, 1, 1) distance = F.abs(an_centers - gt_centers) distance = F.sqrt(F.pow(distance, 2).sum(axis=2)) start = 0 end = 5 overlaps = box_overlap_opr(all_anchors[:, :4], gt_boxes[:, :4]) overlaps = overlaps * valid_mask.unsqueeze(0) default_num = 9 ious_list = [] for l in range(start, end): index = torch.nonzero(all_anchors[:,4].eq(l), as_tuple=False)[:, 0] level_dist = level_dist[index, :].transpose(1, 0) ious = distance[index, :].transpose(1, 0) sorted_index = torch.argsort(ious, 1, descending=False) n = min(default_num, sorted_index.shape[1]) ious = torch.gather(ious, 1, sorted_index[:, :n]).transpose(1, 0) ious_list.append(ious) ious = F.concat(ious_list, axis=0) mean_var = ious.mean(0) std_var = ious.std(0) iou_thresh_per_gt = mean_var + std_var iou_thresh_per_gt = torch.clamp(iou_thresh_per_gt, 0.35) n = iou_thresh_per_gt.shape[0] # limits the anchor centers in the gtboxes N, K = all_anchors.shape[0], gt_boxes.shape[0] anchor_points = an_centers proxies = gt_boxes.unsqueeze(0).repeat(N, 1, 1) l = anchor_points[:, :, 0] - proxies[:, :, 0] r = proxies[:, :, 2] - anchor_points[:, :, 0] t = anchor_points[:, :, 1] - proxies[:, :, 1] b = proxies[:, :, 3] - anchor_points[:, :, 1] is_in_gt = F.stack([l, r, t, b], axis=2) is_in_gt = is_in_gt.min(axis = 2) > 0.1 valid_mask = (overlaps >= iou_thresh_per_gt.unsqueeze(0)) * is_in_gt ious = overlaps * valid_mask argmax_overlaps = torch.argmax(ious, axis=1) max_overlaps = torch.gather(ious, 1, argmax_overlaps.unsqueeze(1)) n = all_anchors.shape[0] labels = -F.ones(n) positive_mask = max_overlaps > 0 negative_mask = max_overlaps < config.rpn_negative_overlap labels = positive_mask + labels * (1 - positive_mask) * (1 - negative_mask) bbox_targets = gt_boxes[argmax_overlaps, :4] bbox_targets = bbox_transform_opr(all_anchors[:, :4], bbox_targets) labels_cat = gt_boxes[argmax_overlaps, 4] labels_cat = labels_cat * (1 - labels.eq(0).astype(np.float32)) labels_cat = labels_cat * (1 - labels.eq(-1).astype(np.float32)) - labels.eq(-1).astype(np.float32) return labels, bbox_targets, labels_cat def rpn_anchor_target_opr(gt_boxes, im_info, anchors): rpn_label_list, rpn_target_boxes_list, iou_thresh_list = [], [], [] for i in range(config.train_batch_per_gpu): rpn_labels, rpn_target_boxes, _ = _anchor_double_target(gt_boxes[i], im_info[i], anchors) rpn_labels = rpn_labels.reshape(-1, 2) c = rpn_target_boxes.shape[1] rpn_target_boxes = rpn_target_boxes.reshape(-1, 2, c) # mask the anchors overlapping with ignore regions ignore_label = mask_anchor_opr(gt_boxes[i], im_info[i], anchors, rpn_labels[:, 0]) rpn_labels = rpn_labels - F.equal(rpn_labels, 0).astype(np.float32) *
F.expand_dims(ignore_label < 0, 1)
megengine.functional.expand_dims
import os import time from megengine.distributed.group import is_distributed import megengine.distributed as dist from megengine.data.dataloader import DataLoader from edit.core.hook import Hook from edit.utils import to_list, is_list_of, get_logger, mkdir_or_exist class EvalIterHook(Hook): """evaluation hook by iteration-based. This hook will regularly perform evaluation in a given interval Args: dataloader (DataLoader): A mge dataloader. interval (int): Evaluation interval. Default: 3000. eval_kwargs (dict): Other eval kwargs. It contains: save_image (bool): Whether to save image. save_path (str): The path to save image. """ def __init__(self, dataloader, **eval_kwargs): if not isinstance(dataloader, DataLoader): raise TypeError('dataloader must be a mge DataLoader, but got {}'.format(type(dataloader))) self.dataloader = dataloader self.eval_kwargs = eval_kwargs self.interval = self.eval_kwargs.pop('interval', 10000) self.save_image = self.eval_kwargs.pop('save_image', False) self.save_path = self.eval_kwargs.pop('save_path', None) self.log_path = self.eval_kwargs.pop('log_path', None) self.multi_process = self.eval_kwargs.pop('multi_process', False) self.ensemble = self.eval_kwargs.pop('ensemble', False) mkdir_or_exist(self.save_path) self.logger = get_logger(name = "EvalIterHook", log_file=self.log_path) # only for rank0 if
is_distributed()
megengine.distributed.group.is_distributed
import os import time from megengine.distributed.group import is_distributed import megengine.distributed as dist from megengine.data.dataloader import DataLoader from edit.core.hook import Hook from edit.utils import to_list, is_list_of, get_logger, mkdir_or_exist class EvalIterHook(Hook): """evaluation hook by iteration-based. This hook will regularly perform evaluation in a given interval Args: dataloader (DataLoader): A mge dataloader. interval (int): Evaluation interval. Default: 3000. eval_kwargs (dict): Other eval kwargs. It contains: save_image (bool): Whether to save image. save_path (str): The path to save image. """ def __init__(self, dataloader, **eval_kwargs): if not isinstance(dataloader, DataLoader): raise TypeError('dataloader must be a mge DataLoader, but got {}'.format(type(dataloader))) self.dataloader = dataloader self.eval_kwargs = eval_kwargs self.interval = self.eval_kwargs.pop('interval', 10000) self.save_image = self.eval_kwargs.pop('save_image', False) self.save_path = self.eval_kwargs.pop('save_path', None) self.log_path = self.eval_kwargs.pop('log_path', None) self.multi_process = self.eval_kwargs.pop('multi_process', False) self.ensemble = self.eval_kwargs.pop('ensemble', False) mkdir_or_exist(self.save_path) self.logger = get_logger(name = "EvalIterHook", log_file=self.log_path) # only for rank0 if is_distributed(): self.local_rank = dist.get_rank() self.nranks = dist.get_world_size() else: self.local_rank = 0 self.nranks = 1 def after_train_iter(self, runner): if not self.every_n_iters(runner, self.interval): return self.logger.info("start to eval for iter: {}".format(runner.iter+1)) save_path = os.path.join(self.save_path, "iter_{}".format(runner.iter+1)) mkdir_or_exist(save_path) results = [] # list of dict if self.multi_process: assert is_distributed(), "when set multiprocess eval, you should use multi process training" raise NotImplementedError("not support multi process for eval now") elif self.local_rank == 0: # 全部交给rank0来处理 for data in self.dataloader: outputs = runner.model.test_step(data, save_image=self.save_image, save_path=save_path, ensemble=self.ensemble) result = runner.model.cal_for_eval(outputs, data) assert isinstance(result, list) results += result self.evaluate(results, runner.iter+1) else: pass if
is_distributed()
megengine.distributed.group.is_distributed
import os import time from megengine.distributed.group import is_distributed import megengine.distributed as dist from megengine.data.dataloader import DataLoader from edit.core.hook import Hook from edit.utils import to_list, is_list_of, get_logger, mkdir_or_exist class EvalIterHook(Hook): """evaluation hook by iteration-based. This hook will regularly perform evaluation in a given interval Args: dataloader (DataLoader): A mge dataloader. interval (int): Evaluation interval. Default: 3000. eval_kwargs (dict): Other eval kwargs. It contains: save_image (bool): Whether to save image. save_path (str): The path to save image. """ def __init__(self, dataloader, **eval_kwargs): if not isinstance(dataloader, DataLoader): raise TypeError('dataloader must be a mge DataLoader, but got {}'.format(type(dataloader))) self.dataloader = dataloader self.eval_kwargs = eval_kwargs self.interval = self.eval_kwargs.pop('interval', 10000) self.save_image = self.eval_kwargs.pop('save_image', False) self.save_path = self.eval_kwargs.pop('save_path', None) self.log_path = self.eval_kwargs.pop('log_path', None) self.multi_process = self.eval_kwargs.pop('multi_process', False) self.ensemble = self.eval_kwargs.pop('ensemble', False) mkdir_or_exist(self.save_path) self.logger = get_logger(name = "EvalIterHook", log_file=self.log_path) # only for rank0 if is_distributed(): self.local_rank =
dist.get_rank()
megengine.distributed.get_rank
import os import time from megengine.distributed.group import is_distributed import megengine.distributed as dist from megengine.data.dataloader import DataLoader from edit.core.hook import Hook from edit.utils import to_list, is_list_of, get_logger, mkdir_or_exist class EvalIterHook(Hook): """evaluation hook by iteration-based. This hook will regularly perform evaluation in a given interval Args: dataloader (DataLoader): A mge dataloader. interval (int): Evaluation interval. Default: 3000. eval_kwargs (dict): Other eval kwargs. It contains: save_image (bool): Whether to save image. save_path (str): The path to save image. """ def __init__(self, dataloader, **eval_kwargs): if not isinstance(dataloader, DataLoader): raise TypeError('dataloader must be a mge DataLoader, but got {}'.format(type(dataloader))) self.dataloader = dataloader self.eval_kwargs = eval_kwargs self.interval = self.eval_kwargs.pop('interval', 10000) self.save_image = self.eval_kwargs.pop('save_image', False) self.save_path = self.eval_kwargs.pop('save_path', None) self.log_path = self.eval_kwargs.pop('log_path', None) self.multi_process = self.eval_kwargs.pop('multi_process', False) self.ensemble = self.eval_kwargs.pop('ensemble', False) mkdir_or_exist(self.save_path) self.logger = get_logger(name = "EvalIterHook", log_file=self.log_path) # only for rank0 if is_distributed(): self.local_rank = dist.get_rank() self.nranks =
dist.get_world_size()
megengine.distributed.get_world_size
import os import time from megengine.distributed.group import is_distributed import megengine.distributed as dist from megengine.data.dataloader import DataLoader from edit.core.hook import Hook from edit.utils import to_list, is_list_of, get_logger, mkdir_or_exist class EvalIterHook(Hook): """evaluation hook by iteration-based. This hook will regularly perform evaluation in a given interval Args: dataloader (DataLoader): A mge dataloader. interval (int): Evaluation interval. Default: 3000. eval_kwargs (dict): Other eval kwargs. It contains: save_image (bool): Whether to save image. save_path (str): The path to save image. """ def __init__(self, dataloader, **eval_kwargs): if not isinstance(dataloader, DataLoader): raise TypeError('dataloader must be a mge DataLoader, but got {}'.format(type(dataloader))) self.dataloader = dataloader self.eval_kwargs = eval_kwargs self.interval = self.eval_kwargs.pop('interval', 10000) self.save_image = self.eval_kwargs.pop('save_image', False) self.save_path = self.eval_kwargs.pop('save_path', None) self.log_path = self.eval_kwargs.pop('log_path', None) self.multi_process = self.eval_kwargs.pop('multi_process', False) self.ensemble = self.eval_kwargs.pop('ensemble', False) mkdir_or_exist(self.save_path) self.logger = get_logger(name = "EvalIterHook", log_file=self.log_path) # only for rank0 if is_distributed(): self.local_rank = dist.get_rank() self.nranks = dist.get_world_size() else: self.local_rank = 0 self.nranks = 1 def after_train_iter(self, runner): if not self.every_n_iters(runner, self.interval): return self.logger.info("start to eval for iter: {}".format(runner.iter+1)) save_path = os.path.join(self.save_path, "iter_{}".format(runner.iter+1)) mkdir_or_exist(save_path) results = [] # list of dict if self.multi_process: assert
is_distributed()
megengine.distributed.group.is_distributed
import os import time from megengine.distributed.group import is_distributed import megengine.distributed as dist from megengine.data.dataloader import DataLoader from edit.core.hook import Hook from edit.utils import to_list, is_list_of, get_logger, mkdir_or_exist class EvalIterHook(Hook): """evaluation hook by iteration-based. This hook will regularly perform evaluation in a given interval Args: dataloader (DataLoader): A mge dataloader. interval (int): Evaluation interval. Default: 3000. eval_kwargs (dict): Other eval kwargs. It contains: save_image (bool): Whether to save image. save_path (str): The path to save image. """ def __init__(self, dataloader, **eval_kwargs): if not isinstance(dataloader, DataLoader): raise TypeError('dataloader must be a mge DataLoader, but got {}'.format(type(dataloader))) self.dataloader = dataloader self.eval_kwargs = eval_kwargs self.interval = self.eval_kwargs.pop('interval', 10000) self.save_image = self.eval_kwargs.pop('save_image', False) self.save_path = self.eval_kwargs.pop('save_path', None) self.log_path = self.eval_kwargs.pop('log_path', None) self.multi_process = self.eval_kwargs.pop('multi_process', False) self.ensemble = self.eval_kwargs.pop('ensemble', False) mkdir_or_exist(self.save_path) self.logger = get_logger(name = "EvalIterHook", log_file=self.log_path) # only for rank0 if is_distributed(): self.local_rank = dist.get_rank() self.nranks = dist.get_world_size() else: self.local_rank = 0 self.nranks = 1 def after_train_iter(self, runner): if not self.every_n_iters(runner, self.interval): return self.logger.info("start to eval for iter: {}".format(runner.iter+1)) save_path = os.path.join(self.save_path, "iter_{}".format(runner.iter+1)) mkdir_or_exist(save_path) results = [] # list of dict if self.multi_process: assert is_distributed(), "when set multiprocess eval, you should use multi process training" raise NotImplementedError("not support multi process for eval now") elif self.local_rank == 0: # 全部交给rank0来处理 for data in self.dataloader: outputs = runner.model.test_step(data, save_image=self.save_image, save_path=save_path, ensemble=self.ensemble) result = runner.model.cal_for_eval(outputs, data) assert isinstance(result, list) results += result self.evaluate(results, runner.iter+1) else: pass if is_distributed():
dist.group_barrier()
megengine.distributed.group_barrier
# -*- coding: utf-8 -*- # MegEngine is Licensed under the Apache License, Version 2.0 (the "License") # # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. # # Unless required by applicable law or agreed to in writing, # software distributed under the License is distributed on an # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. from functools import partial import numpy as np import pytest from utils import opr_test import megengine.functional as F from megengine import jit, tensor def common_test_reduce(opr, ref_opr): data1_shape = (5, 6, 7) data2_shape = (2, 9, 12) data1 = np.random.random(data1_shape).astype(np.float32) data2 = np.random.random(data2_shape).astype(np.float32) cases = [ {"input": data1}, {"input": data2}, {"input": np.array([[[1, 2, np.nan, 4], [8, 6, 5, 2], [2, 3, 4, 5]]])}, ] if opr not in (F.argmin, F.argmax): # test default axis opr_test(cases, opr, ref_fn=ref_opr) # test all axises in range of input shape for axis in range(-3, 3): # test keepdims False opr_test(cases, opr, ref_fn=lambda x: ref_opr(x, axis=axis), axis=axis) # test keepdims True opr_test( cases, opr, ref_fn=lambda x: ref_opr(x, axis=axis, keepdims=True), axis=axis, keepdims=True, ) else: # test defaut axis opr_test(cases, opr, ref_fn=lambda x: ref_opr(x).astype(np.int32)) # test all axises in range of input shape for axis in range(0, 3): opr_test( cases, opr, ref_fn=lambda x: ref_opr(x, axis=axis).astype(np.int32), axis=axis, ) # test negative axis axis = axis - len(data1_shape) opr_test( cases, opr, ref_fn=lambda x: ref_opr(x, axis=axis).astype(np.int32), axis=axis, ) def test_sum(): common_test_reduce(opr=F.sum, ref_opr=np.sum) def test_prod(): common_test_reduce(opr=F.prod, ref_opr=np.prod) def test_mean(): common_test_reduce(opr=F.mean, ref_opr=np.mean) def test_var(): common_test_reduce(opr=F.var, ref_opr=np.var) def test_std(): common_test_reduce(opr=F.std, ref_opr=np.std) def test_min(): common_test_reduce(opr=F.min, ref_opr=np.min) def test_max(): common_test_reduce(opr=F.max, ref_opr=np.max) def test_argmin(): common_test_reduce(opr=F.argmin, ref_opr=np.argmin) def test_argmax(): common_test_reduce(opr=F.argmax, ref_opr=np.argmax) def test_sqrt(): d1_shape = (15,) d2_shape = (25,) d1 = np.random.random(d1_shape).astype(np.float32) d2 = np.random.random(d2_shape).astype(np.float32) cases = [{"input": d1}, {"input": d2}] opr_test(cases, F.sqrt, ref_fn=np.sqrt) def test_sort(): data1_shape = (10, 3) data2_shape = (12, 2) data1 = np.random.random(data1_shape).astype(np.float32) data2 = np.random.random(data2_shape).astype(np.float32) output1 = [np.sort(data1), np.argsort(data1).astype(np.int32)] output2 = [np.sort(data2), np.argsort(data2).astype(np.int32)] cases = [ {"input": data1, "output": output1}, {"input": data2, "output": output2}, ] opr_test(cases, F.sort) @pytest.mark.parametrize("is_symbolic", [None, False, True]) def test_sort_empty(is_symbolic): data_shapes = [ (0,), (10, 0), ] def fn(x): return
F.sort(x)
megengine.functional.sort
# -*- coding: utf-8 -*- # MegEngine is Licensed under the Apache License, Version 2.0 (the "License") # # Copyright (c) 2014-2021 Megvii Inc. All rights reserved. # # Unless required by applicable law or agreed to in writing, # software distributed under the License is distributed on an # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. from functools import partial import numpy as np import pytest from utils import opr_test import megengine.functional as F from megengine import jit, tensor def common_test_reduce(opr, ref_opr): data1_shape = (5, 6, 7) data2_shape = (2, 9, 12) data1 = np.random.random(data1_shape).astype(np.float32) data2 = np.random.random(data2_shape).astype(np.float32) cases = [ {"input": data1}, {"input": data2}, {"input": np.array([[[1, 2, np.nan, 4], [8, 6, 5, 2], [2, 3, 4, 5]]])}, ] if opr not in (F.argmin, F.argmax): # test default axis opr_test(cases, opr, ref_fn=ref_opr) # test all axises in range of input shape for axis in range(-3, 3): # test keepdims False opr_test(cases, opr, ref_fn=lambda x: ref_opr(x, axis=axis), axis=axis) # test keepdims True opr_test( cases, opr, ref_fn=lambda x: ref_opr(x, axis=axis, keepdims=True), axis=axis, keepdims=True, ) else: # test defaut axis opr_test(cases, opr, ref_fn=lambda x: ref_opr(x).astype(np.int32)) # test all axises in range of input shape for axis in range(0, 3): opr_test( cases, opr, ref_fn=lambda x: ref_opr(x, axis=axis).astype(np.int32), axis=axis, ) # test negative axis axis = axis - len(data1_shape) opr_test( cases, opr, ref_fn=lambda x: ref_opr(x, axis=axis).astype(np.int32), axis=axis, ) def test_sum(): common_test_reduce(opr=F.sum, ref_opr=np.sum) def test_prod(): common_test_reduce(opr=F.prod, ref_opr=np.prod) def test_mean(): common_test_reduce(opr=F.mean, ref_opr=np.mean) def test_var(): common_test_reduce(opr=F.var, ref_opr=np.var) def test_std(): common_test_reduce(opr=F.std, ref_opr=np.std) def test_min(): common_test_reduce(opr=F.min, ref_opr=np.min) def test_max(): common_test_reduce(opr=F.max, ref_opr=np.max) def test_argmin(): common_test_reduce(opr=F.argmin, ref_opr=np.argmin) def test_argmax(): common_test_reduce(opr=F.argmax, ref_opr=np.argmax) def test_sqrt(): d1_shape = (15,) d2_shape = (25,) d1 = np.random.random(d1_shape).astype(np.float32) d2 = np.random.random(d2_shape).astype(np.float32) cases = [{"input": d1}, {"input": d2}] opr_test(cases, F.sqrt, ref_fn=np.sqrt) def test_sort(): data1_shape = (10, 3) data2_shape = (12, 2) data1 = np.random.random(data1_shape).astype(np.float32) data2 = np.random.random(data2_shape).astype(np.float32) output1 = [np.sort(data1), np.argsort(data1).astype(np.int32)] output2 = [np.sort(data2), np.argsort(data2).astype(np.int32)] cases = [ {"input": data1, "output": output1}, {"input": data2, "output": output2}, ] opr_test(cases, F.sort) @pytest.mark.parametrize("is_symbolic", [None, False, True]) def test_sort_empty(is_symbolic): data_shapes = [ (0,), (10, 0), ] def fn(x): return F.sort(x) for shape in data_shapes: if is_symbolic is not None: fn_ = jit.trace(symbolic=is_symbolic)(fn) else: fn_ = fn data = np.random.random(shape).astype(np.float32) for _ in range(3): outs = fn_(tensor(data)) ref_outs = (np.sort(data), np.argsort(data)) assert len(ref_outs) == len(outs) for i in range(len(outs)): np.testing.assert_equal(outs[i].numpy(), ref_outs[i]) if is_symbolic is None: break def test_normalize(): cases = [ {"input": np.random.random((2, 3, 12, 12)).astype(np.float32)} for i in range(2) ] def np_normalize(x, p=2, axis=None, eps=1e-12): if axis is None: norm = np.sum(x ** p) ** (1.0 / p) else: norm = np.sum(x ** p, axis=axis, keepdims=True) ** (1.0 / p) return x / np.clip(norm, a_min=eps, a_max=np.inf) # # Test L-2 norm along all dimensions # opr_test(cases, F.normalize, ref_fn=np_normalize) # # Test L-1 norm along all dimensions # opr_test(cases, partial(F.normalize, p=1), ref_fn=partial(np_normalize, p=1)) # Test L-2 norm along the second dimension opr_test(cases, partial(F.normalize, axis=1), ref_fn=partial(np_normalize, axis=1)) # Test some norm == 0 cases[0]["input"][0, 0, 0, :] = 0 cases[1]["input"][0, 0, 0, :] = 0 opr_test(cases, partial(F.normalize, axis=3), ref_fn=partial(np_normalize, axis=3)) def test_sum_neg_axis(): shape = (2, 3) data = np.random.random(shape).astype(np.float32) for axis in (-1, -2, (-2, 1), (-1, 0)): get = F.sum(tensor(data), axis=axis) ref = np.sum(data, axis=axis) np.testing.assert_allclose(get.numpy(), ref, rtol=1e-6) with pytest.raises(AssertionError): F.sum(tensor(data), axis=(-1, 1)) def test_non_finite(): shape = (32, 3, 32, 32) data1 = np.random.random(shape).astype(np.float32) data2 = np.random.random(shape).astype(np.float32) rst = F.math._check_non_finite([tensor(data1), tensor(data2)]) np.testing.assert_equal(rst.numpy(), [0]) data2[0][0][0][0] = float("inf") rst = F.math._check_non_finite([tensor(data1), tensor(data2)]) np.testing.assert_equal(rst.numpy(), [1]) data2[0][0][0][0] = float("nan") rst = F.math._check_non_finite([tensor(data1), tensor(data2)]) np.testing.assert_equal(rst.numpy(), [1]) @pytest.mark.parametrize("descending", [True, False]) @pytest.mark.parametrize("sorted", [True, False]) @pytest.mark.parametrize("inp1d", [True, False]) @pytest.mark.parametrize("kth_only", [True, False]) def test_topk(descending, sorted, inp1d, kth_only): k = 3 if inp1d: data = np.random.permutation(7) else: data = np.random.permutation(5 * 7).reshape(5, 7) data = data.astype(np.int32) def np_sort(x): if descending: return np.sort(x)[..., ::-1] return np.sort(x) res = F.topk(
tensor(data)
megengine.tensor