vill-lab/2022-tip-hcga

Human Co-Parsing Guided Alignment for Occluded Person Re-identification（IEEE T-IP 23）

Human Co-Parsing Guided Alignment for Occluded Person Re-identification (IEEE T-IP 2023)

This is the Pytorch implementation of the paper. More information about the paper is in here.

We propose a novel Human Co-parsing Guided Alignment (HCGA) framework that alternately trains the human co-parsing network and the ReID network, where the human co-paring network is trained in a weakly supervised manner to obtain paring results without any extra annotation.

Installation

Clone this repository and install its requirements.

conda create -n hcga
conda activate hcga
conda install -c pytorch faiss-gpu

# For RTX 3090 and Tesla A100, we use CUDA 11.1.
pip install torch==1.10.1+cu111 torchvision==0.11.2+cu111 torchaudio==0.10.1 -f https://download.pytorch.org/whl/torch_stable.html

git clone https://github.com/Vill-Lab/2022-TIP-HCGA
cd 2022-TIP-HCGA
pip install -r requirements.txt

Reproducing the results

1. Prepare the ReID datasets

2. Download the pre-trained HRNet32 on ImageNet from Model, code: r1o2.

3. The following command will train HCGA on Occluded-Duke.

bash HCGA-OD.sh

Note: use GPU multi-process need large memory and GPU Memory (We use RTX 3090 with 24GB).

4. Test

bash Test-OD.sh

Reference

We hope that this technique will benefit more computer vision related applications and inspire more works. If you find this technique and repository useful, please cite the paper. Thanks!

@article{hcga23tip,
  author={Dou, Shuguang and Zhao, Cairong and Jiang, Xinyang and Zhang, Shanshan and Zheng, Wei-Shi and Zuo, Wangmeng},
  journal={IEEE Transactions on Image Processing}, 
  title={Human Co-Parsing Guided Alignment for Occluded Person Re-Identification}, 
  year={2023},
  volume={32},
  pages={458-470},
  doi={10.1109/TIP.2022.3229639}}