Camera-Conditioned Stable Feature Generation for Isolated Camera Supervised Person Re-IDentification （CVPR'22）

This repository is Pytorch code for our proposed Camera-Conditioned Stable Feature Generation (CCSFG).

Paper link: https://arxiv.org/abs/2203.15210

Environment

The code is based on fastreid. See INSTALL.md.

Dataset Preparation

1. Download Market-1501 and MSMT17.
2. Split Market-1501 and MSMT17 to Market-SCT and MSMT-SCT according to the file names in the market_sct.txt and msmt_sct.txt.
3. vim fastreid/data/build.py change the _root to your own data folder.
4. Make new directories in datasets and organize them as follows:

+-- datasets
|   +-- market
|       +-- bounding_box_train_sct
|       +-- query
|       +-- boudning_box_test
|   +-- msmt
|       +-- bounding_box_train_sct
|       +-- query
|       +-- boudning_box_test

Train and test

To train and test the model, you can use following command:

CUDA_VISIBLE_DEVICES=0 python tools/train_net.py --config-file configs/Msmt/AGW_R50.yml

If you only want to test the model, you can download our model Google Drive or Baidu Drive (Code:0000) to logs/ and use following command:

CUDA_VISIBLE_DEVICES=0 python tools/train_net.py --config-file configs/Msmt/AGW_R50.yml --eval-only  MODEL.WEIGHTS logs/msmt.pth

Citation

If you find this code useful, please kindly cite the following paper:

@article{wu2022camera,
  title={Camera-Conditioned Stable Feature Generation for Isolated Camera Supervised Person Re-IDentification},
  author={Wu, Chao and Ge, Wenhang and Wu, Ancong and Chang, Xiaobin},
  booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2022}
}