Co-Attention Aligned Mutual Cross-Attention for Cloth-Changing Person Re-Identification

Official PyTorch implementation of "Co-Attention Aligned Mutual Cross-Attention for Cloth-Changing Person Re-Identification". (ACCV 2022 Oral)

Qizao Wang, Xuelin Qian, Yanwei Fu, Xiangyang Xue

Fudan University

Getting Started

Environment

• Python == 3.8
• PyTorch == 1.12.1

Prepare Data

Please download cloth-changing person Re-ID datasets and place them in any path DATASET_ROOT. Take Celeb-reID as an example:

DATASET_ROOT
	└─ Celeb-reID
		├── train
		├── query
		└── gallery

Training

python main.py --gpu_devices 0 --pose_net_path POSE_NET_PATH --dataset celeb --dataset_root DATASET_ROOT --dataset_filename Celeb-reID --save_dir SAVE_DIR --save_checkpoint

--pose_net_path : replace POSE_NET_PATH with the path of pretrained HRNet weights (download here)

--dataset_root : replace DATASET_ROOT with your dataset root path

--save_dir: replace SAVE_DIR with the path to save log files and checkpoints

Evaluation

python main.py --gpu_devices 0 --pose_net_path POSE_NET_PATH --dataset celeb --dataset_root DATASET_ROOT --dataset_filename Celeb-reID --resume RESUME_PATH --save_dir SAVE_DIR --evaluate

--resume: replace RESUME_PATH with the path of the saved checkpoint

Results

• Celeb-reID

Backbone	Rank-1	Rank-5	mAP
ResNet-50	57.5	71.5	12.3

• LTCC

Backbone	Setting	Rank-1	mAP
ResNet-50	Cloth-Changing	36.0	15.4
ResNet-50	Standard	73.2	35.3

You can achieve similar results with released code.

For implementation simplicity, the code in the repository does not support the cloth-changing setting on the LTCC dataset. Please refer to our latest work, FIRe-CCReID or CSSC-CCReID, which includes additional functionality and code for other datasets.

Citation

Please cite the following paper in your publications if it helps your research:

@inproceedings{wang2022co,
  title={Co-attention aligned mutual cross-attention for cloth-changing person re-identification},
  author={Wang, Qizao and Qian, Xuelin and Fu, Yanwei and Xue, Xiangyang},
  booktitle={Proceedings of the Asian Conference on Computer Vision},
  pages={2270--2288},
  year={2022}
}

Contact

Any questions or discussions are welcome!

Qizao Wang ([email protected])