ABL-HD

🌟 New! ABLkit released: A toolkit for Abductive Learning with high flexibility, easy-to-use interface, and optimized performance. Welcome to try it out!🚀

KESAR: Knowledge Enhanced Historical Document Sagmentation and Recognition

This is the code for KESAR, an Abductive Learning-based model training method for historical document segmentation and recognition.

Publication

"Knowledge-Enhanced Historical Document Segmentation and Recognition". En-Hao Gao, Yu-Xuan Huang, Wen-Chao Hu, Xin-Hao Zhu, Wang-Zhou Dai. In: Proceedings of the 38th AAAI Conference on Artificial Intelligence (AAAI’24), Vancouver, Canada, 2024, pp.8409-8416 (Oral).[paper]

Environment

conda create -n kesar python=3.7
conda activate kesar
pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113
pip install onnx onnxruntime-gpu python-opencv scipy scikit-image shapely tqdm segmentation_models_pytorch

Inference

You need to change 3 paths before inference.

• Segmentation model path, Line 19 ~ 22 in TestModel.py
• Word model path, Line 30 ~ 31 in TestModel.py
• Image path, Line 116 in TestModel.py

After changing these paths, you can run the following commond to conduct inference:

python TestModel.py

Results will be saved into the "outputs/" folder.

Citation

@inproceedings{KESAR2024Gao,
  author     = {Gao, En-Hao and Huang, Yu-Xuan and Hu, Wen-Chao and Zhu, Xin-Hao and Dai, Wang-Zhou},
  title      = {Knowledge-Enhanced Historical Document Segmentation and Recognition},
  booktitle  = {Proceedings of the 38th AAAI Conference on Artificial Intelligence (AAAI'24)},
  pages      = {8409--8416},
  year       = {2024}
}