Skip to content

Latest commit

 

History

History
81 lines (64 loc) · 3.75 KB

README.md

File metadata and controls

81 lines (64 loc) · 3.75 KB
Raw

GeoLayoutLM: Geometric Pre-training for Visual Information Extraction

The official PyTorch implementation of GeoLayoutLM (CVPR 2023 highlight).

Paper

GeoLayoutLM is a multi-modal framework for Visual Information Extraction (VIE, including SER and RE), which incorporates the novel geometric pre-training. Additionally, novel relation heads, which are pre-trained by the geometric pre-training tasks and fine-tuned for RE, are designed to enrich and enhance the feature representation. GeoLayoutLM achieves highly competitive scores in the SER task, and significantly outperforms the previous state-of-the-arts for RE.

Geometric pre-training

Environment

The dependencies are listed in requirements.txt. Please install a proper torch version matching your cuda version first.

pip install -r requirements.txt

Model Checkpoints

We released the pre-trained model for downstream fine-tuning. Also, we provided SER and RE models fine-tuned on FUNSD.

Pre-trained model SER model RE model
Download LINK LINK LINK
F1 - 92.32 89.23

Note that the training with the vision module causes unstable final performance, i.e., multiple independent experiments will have diffenrent F1 scores.

Fine-tuning

Preprocess the data

  • FUNSD

FUNSD is a dataset for form understanding. It is widely used in the VIE task.

cd preprocess/funsd_el/
python preprocess.py

Fine-tune and Evaluate

CUDA_VISIBLE_DEVICES=0 python train.py --config=configs/finetune_funsd.yaml
CUDA_VISIBLE_DEVICES=0 python evaluate.py --config=configs/finetune_funsd.yaml [--pretrained_model_file=path/to/xx.pt]

Multi-lingual base model

We also released a base model pre-trained on Chinese and English documents. Refer to modelscope for more details.

Acknowledgments

We implemented the fine-tuning based on the code of BROS.

Citation

Please cite our paper if the work helps you.

@article{cvpr2023geolayoutlm,
  title={GeoLayoutLM: Geometric Pre-training for Visual Information Extraction},
  author={Chuwei Luo and Changxu Cheng and Qi Zheng and Cong Yao},
  journal={2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2023}
}

License

Copyright 2023-present Alibaba Group.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.