Hope ERNIE-Layout can play withe Hugging face, not only support for paddle paddle.Thanks.

#1
by luojiluoji - opened

Hope ERNIE-Layout can play withe Hugging face, not only support for paddle paddle.
Hugging face is more universality.
Thanks.

PaddlePaddle org

Stay tuned ๐Ÿ˜‰

PaddlePaddle org

@luojiluoji check out our latest work https://huggingface.co/spaces/PaddlePaddle/UIE-X! The underlying model is open-sourced on the hub too.

Is Ernie-layout open sourced? really impressive, tried it on bond term sheets, very accurate.

PaddlePaddle org

hi @nickmuchi . The inference model of docprompt_v1 is open sourced (not on HF Hub yet). You use it through our Taskflow.

@sijunhe how can I get bounding boxes using Taskflow? I need the bounding boxes so I can label the original image as displayed in this Space.

PaddlePaddle org
โ€ข
edited Jun 1, 2023

@sijunhe how can I get bounding boxes using Taskflow? I need the bounding boxes so I can label the original image as displayed in this Space.

@nikolab
The OCR results are generated here as part of the preprocessing and passed into the model here. Currently it's not part of the output, but you can modify it somewhere around here

@sijunhe how can I get bounding boxes using Taskflow? I need the bounding boxes so I can label the original image as displayed in this Space.

@nikolab
The OCR results are generated here as part of the preprocessing and passed into the model here. Currently it's not part of the output, but you can modify it somewhere around here

Thanks @sijunhe for the help! I managed to extract bounding boxes and display them in the image.

What I am doing is matching the model's output with the OCR results and extracting the bounding boxes from the appropriate OCR item. Is it guaranteed that the model's output (if not null) is chosen from the OCR results, depending on the input question (prompt)?

Sign up or log in to comment