Skip to content

Latest commit

 

History

History
 
 

lgpma

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
configs
configs
 
 
tools
tools
 
 
vis
vis
 
 
dist_train.sh
dist_train.sh
 
 
readme.md
readme.md
 
 

readme.md

LGPMA

This code repository contains the implementations of the paper LGPMA: Complicated Table Structure Recognition with Local and Global Pyramid Mask Alignment (ICDAR 2021).

Preparing Dataset

Original images can be downloaded from: pubtabnet.

The test datalist and the example of formatted training datalist can be found in demo/table_recognition/datalist/

The whole formatted training datalist can be downloaded from: PubTabNet_train_datalist_all.json [extraction code is 7gto].

Training

Modified the paths of "ann_file", "img_prefix", "pretrained_model" and "work_space" in the config files demo/table_recognition/lgpma/config/lgpma_pub.py.

Run the following bash command in the command line,

cd $DAVAR_LAB_OCR_ROOT$/demo/table_recognition/lgpma/
bash dist_train.sh

Offline Inference and Evaluation

We provide a demo of forward inference and evaluation on PubTabNet dataset. You can modify the paths (savepath, config_file, checkpoint_file) in test script, and start testing:

python test_pub.py

Some visualization of detection results are shown:

./vis/PMC2871264_002_00.png

./vis/PMC3160368_005_00.png

./vis/PMC3250619_005_01.png

./vis/PMC3551656_004_00.png

./vis/PMC3568059_003_00.png

./vis/PMC3824233_004_00.png

The offline evaluation tool can be found in demo/table_recognition/lgpma/tools/eval_pub/

Trained Model Download

All of the models are re-implemented and well trained in the based on the opensourced framework mmdetection. So, the results might be slightly different from reported results.

Results on various datasets and trained models download:

Dataset Test Scale TEDS-struc Links
PubTabNet(reported) L-768 96.7
PubTabNet 1.5x 96.7 config, pth (Access Code: gygm)

The release model only contains structure-level result. You may use the text recognition module for the complete result.

The Trained Model on dataset SciTSR and ICDAR 2013 will release soon.

Note: Models are stored in BaiduYunPan, and can also be downloaded from Google Drive

Citation

If you find this repository is helpful to your research, please feel free to cite us:

@inproceedings{qiao2021icdar21,
  title={LGPMA: Complicated Table Structure Recognition with Local and Global Pyramid Mask Alignment},
  author={Qiao, Liang and Li, Zaisheng and Cheng, Zhanzhan and Zhang, Peng and Pu, Shiliang and Niu, Yi and Ren, Wenqi and Tan, Wenming and Wu, Fei},
  booktitle={Document Analysis and Recognition-ICDAR 2021, 16th International Conference, Lausanne, Switzerland, September 5–10, 2021, Proceedings, Part I},
  pages={99-114},
  year={2021}
}

License

This project is released under the Apache 2.0 license

Copyright

If there is any suggestion and problem, please feel free to contact the author with qiaoliang6@hikvision.com, lizaisheng@hikvision.com or chengzhanzhan@hikvision.com.