Camera-aware Proxies for Unsupervised Person Re-Identification

The official implementation for Camera-aware Proxies for Unsupervised Person Re-Identification, which is accepted by AAAI 2021. CAP (Camera-Aware Proxies) achieves state-of-the-art performance on pure unsupervised person re-ID task. It can also be applied to unsupervised vehicle re-ID with competitive performance.

Preparation

Requirements: Pytorch>=1.1.0 and python>=3.6

1. install pytorch

2. Download re-ID dataset

   • Market-1501
   • DukeMTMC-reID
   • MSMT17
   • VeRi-776

3. Put the data under the dataset directory. Training, query and test sub-folder should named as bounding_box_train, query, bounding_box_test, respectively.

Training and test model for unsupervised re-ID

# train CAP model on Market-1501
CUDA_VISIBLE_DEVICES=0 python train_cap.py --target 'Market1501' --data_dir '/folder/to/dataset' --logs_dir 'Market_logs'

# test model on Market-1501
CUDA_VISIBLE_DEVICES=0 python train_cap.py --target 'Market1501' --data_dir '/folder/to/dataset' --logs_dir 'Market_logs' --evaluate True --load_ckpt 'trained_model_name.pth'

Results

The performance of CAP on Vehicle re-ID dataset VeRi-776:

Rank-1 (%)	mAP (%)
87.0	40.6

Citation

If you find this work useful in your research, please cite the following paper:

@inproceedings{Wang2021camawareproxies,
  title={Camera-aware Proxies for Unsupervised Person Re-Identification},
  author={Menglin Wang and Baisheng Lai and Jianqiang Huang and Xiaojin Gong and Xian-Sheng Hua},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)},
  year={2021},
}