/IIA

Improving Person Re-Identification With Iterative Impression Aggregation (IIA)

Primary LanguagePythonMIT LicenseMIT

Improving Person Re-Identification With Iterative Impression Aggregation (IIA)

The repository is for our TIP paper Improving Person Re-Identification With Iterative Impression Aggregation.

framework

Requirements

  • torch >= 1.0

Extracted Features

The MGN features we used can be download at MGN.

How to use

Please follow the usage of our demo.py script. If you have downloaded our extracted features, you may run like the following:

### online 
python demo.py --dname market --fpath mgn_market.pkl
### offline
python demo.py --dname market --fpath mgn_market.pkl --off

This (IIA_bas) will give 93.73 mAP and 89.60 mAP for Market-1501 and DukeMTMC respectively, which are slightly higher than paper reported.

Citation

If you find this code useful for your research, please cite our paper.

@article{fu2020improving,
  title={Improving Person Re-Identification With Iterative Impression Aggregation},
  author={Fu, Dengpan and Xin, Bo and Wang, Jingdong and Chen, Dongdong and Bao, Jianmin and Hua, Gang and Li, Houqiang},
  journal={IEEE Transactions on Image Processing},
  volume={29},
  pages={9559--9571},
  year={2020},
  publisher={IEEE}
}