search

mdv3101 / CDeCNet

CDeC-Net: Composite Deformable Cascade Network for Table Detection in Document Images
MIT License
132 stars 33 forks source link

Testing the model on IIIT-AR-13K - no detected tables. #8

Closed syrinecheriaa closed 3 years ago

syrinecheriaa commented 3 years ago

I am interesting in using CDeCNet for detecting tables in annual reports. Thus, I tried to use your base model. I downloded the checkpoint that you shared (the file name is epoch_50.pth). After that I tried to use it with image from the dataset IIIT-AR-13K. I used mmdet.apis as suggested in the scrip https://github.com/mdv3101/CDeCNet/blob/master/demo/image_demo.py. I used the config file https://github.com/mdv3101/CDeCNet/blob/master/configs/cascade_rcnn/db_cascade_mask_rcnn_x101_32x4d_fpn_1x_coco.py. The script runs fine but no tables are detected ie the result returned by the command inference_detector(model, img) is empty. I am wondering if I am using the wrong checkpoint file or the wrong config. Thanks.

syrinecheriaa commented 3 years ago

It is clear that the checkpoint and the config file that I used do not match (there are some issues in the layers size). Could you please share the config file that match the base model?

mdv3101 commented 3 years ago

Hi @syrinecheriaa , Kindly use the following config file: configs/dcn/db_cascade_mask_rcnn_x101_fpn_dconv_c3-c5_1x_coco.py

syrinecheriaa commented 3 years ago

Thank you @mdv3101 .