Fix SAM runtime error

app.py

@@ -76,48 +76,6 @@ def resize_inputs(image_input, crop_size):
     results = add_margin(ref_img_, size=256)
     return results
 
-# def generate(model, sample_steps, batch_view_num, sample_num, cfg_scale, seed, image_input, elevation_input):
-#     if deployed:
-#         assert isinstance(model, SyncMultiviewDiffusion)
-#         seed=int(seed)
-#         torch.random.manual_seed(seed)
-#         np.random.seed(seed)
-
-#         # prepare data
-#         image_input = np.asarray(image_input)
-#         image_input = image_input.astype(np.float32) / 255.0
-#         alpha_values = image_input[:,:, 3:]
-#         image_input[:, :, :3] = alpha_values * image_input[:,:, :3] + 1 - alpha_values # white background
-#         image_input = image_input[:, :, :3] * 2.0 - 1.0
-#         image_input = torch.from_numpy(image_input.astype(np.float32))
-#         elevation_input = torch.from_numpy(np.asarray([np.deg2rad(elevation_input)], np.float32))
-#         data = {"input_image": image_input, "input_elevation": elevation_input}
-#         for k, v in data.items():
-#             if deployed:
-#                 data[k] = v.unsqueeze(0).cuda()
-#             else:
-#                 data[k] = v.unsqueeze(0)
-#             data[k] = torch.repeat_interleave(data[k], sample_num, dim=0)
-
-#         if deployed:
-#             sampler = SyncDDIMSampler(model, sample_steps)
-#             x_sample = model.sample(sampler, data, cfg_scale, batch_view_num)
-#         else:
-#             x_sample = torch.zeros(sample_num, 16, 3, 256, 256)
-
-#         B, N, _, H, W = x_sample.shape
-#         x_sample = (torch.clamp(x_sample,max=1.0,min=-1.0) + 1) * 0.5
-#         x_sample = x_sample.permute(0,1,3,4,2).cpu().numpy() * 255
-#         x_sample = x_sample.astype(np.uint8)
-
-#         results = []
-#         for bi in range(B):
-#             results.append(np.concatenate([x_sample[bi,ni] for ni in range(N)], 1))
-#         results = np.concatenate(results, 0)
-#         return Image.fromarray(results)
-#     else:
-#         return Image.fromarray(np.zeros([sample_num*256,16*256,3],np.uint8))
-
 def generate(pipe, image_input, azimuth):
     target_index = round(azimuth % 360 / 22.5)
     output = pipe(conditioning_image=image_input)
@@ -163,12 +121,6 @@ def load_model(cfg,ckpt,strict=True):
     return model
 
 def run_demo():
-    # # device = f"cuda:0" if torch.cuda.is_available() else "cpu"
-    # # models = None # init_model(device, os.path.join(code_dir, ckpt))
-    # cfg = 'configs/syncdreamer.yaml'
-    # ckpt = 'ckpt/syncdreamer-pretrain.ckpt'
-    # config = OmegaConf.load(cfg)
-    # # model = None
 
     if deployed:
         controlnet = ControlNetModelSync.from_pretrained('controlnet_ckpt', torch_dtype=torch.float32, use_safetensors=True)
@@ -182,33 +134,25 @@ def run_demo():
         )
         pipe.to('cuda', dtype=torch.float32)
 
-    # if deployed:
-    #     model = instantiate_from_config(config.model)
-    #     print(f'loading model from {ckpt} ...')
-    #     ckpt = torch.load(ckpt,map_location='cpu')
-    #     model.load_state_dict(ckpt['state_dict'], strict=True)
-    #     model = model.cuda().eval()
-    #     del ckpt
         mask_predictor = sam_init()
         removal = BackgroundRemoval()
     else:
-        # model = None
-        # mask_predictor = None
-        # removal = None
+        mask_predictor = None
+        removal = None
         controlnet = None
         dreamer = None
         pipe = None
 
     # NOTE: Examples must match inputs
     examples_full = [
-        ['hf_demo/examples/monkey.png',30,200],
-        ['hf_demo/examples/cat.png',30,200],
-        ['hf_demo/examples/crab.png',30,200],
-        ['hf_demo/examples/elephant.png',30,200],
-        ['hf_demo/examples/flower.png',0,200],
-        ['hf_demo/examples/forest.png',30,200],
-        ['hf_demo/examples/teapot.png',20,200],
-        ['hf_demo/examples/basket.png',30,200],
+        ['hf_demo/examples/monkey.png',200],
+        ['hf_demo/examples/cat.png',200],
+        ['hf_demo/examples/crab.png',200],
+        ['hf_demo/examples/elephant.png',200],
+        ['hf_demo/examples/flower.png',200],
+        ['hf_demo/examples/forest.png',200],
+        ['hf_demo/examples/teapot.png',200],
+        ['hf_demo/examples/basket.png',200],
     ]
 
     image_block = gr.Image(type='pil', image_mode='RGBA', height=256, label='Input image', tool=None, interactive=True)
@@ -220,16 +164,14 @@ def run_demo():
         with gr.Row():
             with gr.Column(scale=1):
                 gr.Markdown('# ' + _TITLE)
-            # with gr.Column(scale=0):
-            #     gr.DuplicateButton(value='Duplicate Space for private use', elem_id='duplicate-button')
         gr.Markdown(_DESCRIPTION)
 
         with gr.Row(variant='panel'):
             with gr.Column(scale=1.2):
                 gr.Examples(
                     examples=examples_full,  # NOTE: elements must match inputs list!
-                    inputs=[image_block, azimuth, crop_size],
-                    outputs=[image_block, azimuth, crop_size],
+                    inputs=[image_block, crop_size],
+                    outputs=[image_block, crop_size],
                     cache_examples=False,
                     label='Examples (click one of the images below to start)',
                     examples_per_page=5,
@@ -244,7 +186,6 @@ def run_demo():
             with gr.Column(scale=0.8):
                 sam_block = gr.Image(type='pil', image_mode='RGBA', label="SAM output", height=256, interactive=False)
                 crop_size.render()
-                # crop_btn = gr.Button('Crop it', variant='primary', interactive=True)
                 fig1 = gr.Image(value=Image.open('assets/elevation.jpg'), type='pil', image_mode='RGB', height=256, show_label=False, tool=None, interactive=False)
 
             with gr.Column(scale=0.8):
@@ -278,7 +219,6 @@ def run_demo():
         # crop_btn.click(fn=resize_inputs, inputs=[sam_block, crop_size], outputs=[input_block], queue=False)\
         #                .success(fn=partial(update_guide, _USER_GUIDE2), outputs=[guide_text], queue=False)
 
-        # azimuth.render()
         run_btn.click(partial(generate, pipe), inputs=[input_block, azimuth], outputs=[output_block], queue=True)\
                .success(fn=partial(update_guide, _USER_GUIDE3), outputs=[guide_text], queue=False)