[Question] about Table detection

[phamkhactu]

Thanks for excellence repo.

I have read the tutorial for table recognition, but in tutorial only provide the table recognition. I know if want to recognize table(OCR), you must detect region of table. I can not configs to get only bounding box(top-left, bottom-right) for table. I am very happy for your help to get bounding box the table. Thanks in advance.

[johntoma]

You need to use PPStructure Layout Analysis to detect tables.

https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/ppstructure/docs/quickstart_en.md

https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/ppstructure/docs/models_list_en.md

[phamkhactu]

You need to use PPStructure Layout Analysis to detect tables.

https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/ppstructure/docs/quickstart_en.md

https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/ppstructure/docs/models_list_en.md

Thanks for your pointing out, I will try it.

[vani-mcm]

Hii.. I have a problem regarding the output of layout analysis. Even if I cropped my roi(table area), the layout analysis output is showing it as figure, not as table. What could be the reason? Any idea..

[tranthuhoai3786]

@vani-mcm i get the same issue, do you have any solutions?

[phamkhactu]

Hii.. I have a problem regarding the output of layout analysis. Even if I cropped my roi(table area), the layout analysis output is showing it as figure, not as table. What could be the reason? Any idea..

@vani-mcm cc @johntoma Yes, I have the same problem in some casese. However, I checked again, it gave me Bounding box for table not good. It means that roi not concentrate on table(IOU), almost it gave [0,0 , w,h] for bbox

[wwolfyy]

@vani-mcm, @tranthuhoai3786 , the layout models only detect the areas where tables are. To parse the tables you'd need to look at: https://github.com/PaddlePaddle/PaddleOCR/blob/main/ppstructure/table/README.md