An example using the proposed benchmark suite

Introduction

The majority of this project was done during the internship of the first author in the company. By the company's legal policy, only part of the code and dataset can be released. This repo provides a demo of how to use the proposed benchmark dataset.

The code is heavily developed based on this repo, which is based on the faster-rcnn repository. Refer these two repos for environment setup.

Prepare the data

• Download the "Chs" and "PubMed" datasets here. The datasets have been organized in the "Pascal_VOC" format.

• Write the path in __C.VGG_PATH and __C.RESNET_PATH at lib/model/utils/config.py.

Train

• Adaptation from "PubMed" to "Chs"

CUDA_VISIBLE_DEVICES=$GPU_ID python trainval_net_global_local.py --cuda --net res101 --dataset doc_med --dataset_t doc_chs_median --save_dir ./models/ckpts/da_med2chsmed_ps_lr3ss8 --stdout_file da_med2chsmed_ps_lr3ss8 --lr 1e-3 --lr_decay_step 8

• Adaptation from "PubMed" to "Chs"

CUDA_VISIBLE_DEVICES=$GPU_ID python trainval_net_global_local.py --cuda --net res101 --dataset doc_chs_median --dataset_t doc_med --save_dir ./models/ckpts/da_chsmed2med_ps_lr3ss8 --stdout_file da_med2chsmed_ps_lr3ss8 --lr 1e-3 --lr_decay_step 8

Citation

Please cite the following reference if you find this repo helps your research.

@inproceedings{li2020cross,
  title={Cross-Domain Document Object Detection: Benchmark Suite and Method},
  author={Li, Kai and Wigington, Curtis and Tensmeyer, Chris and Zhao, Handong and Barmpalios, Nikolaos and Morariu, Vlad I and Manjunatha, Varun and Sun, Tong and Fu, Yun},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={12915--12924},
  year={2020}
}