ICPR2020 (Oral) CANU-MMT: A Conditional Adversarial Network for Unsupervised person Re-IDentification for MMT
This is a github portal to the implementation of CANU-ReID: A Conditional Adversarial Network for Unsupervised person Re-IDentification, ICPR 2020 (Oral), based on Mutual Mean-Teaching: Pseudo Label Refinery for Unsupervised Domain Adaptation on Person Re-identification.
Please visit https://gitlab.inria.fr/gdelorme/CANU_REID_MMT to the complete code implemented mainly by Guillaume and partially by Yihong. Due to licence issues, we currently only release code for CANU-MMT: we will update the code for SSG asap.
If you like the work, please star the project and cite:
@inproceedings{delorme2021canu,
title={CANU-ReID: A Conditional Adversarial Network for Unsupervised person Re-IDentification},
author={Delorme, Guillaume and Xu, Yihong and Lathuili{\`e}re, St{\'e}phane and Horaud, Radu and Alameda-Pineda, Xavier},
booktitle={International Conference on Pattern Recognition},
year={2021}
}
Weights are available here.
| SRC --> TGT | Adaptation by MMT(dbscan) | Adaptation by Adv+MMT | Adaptation by CANU+MMT | |||
|---|---|---|---|---|---|---|
| Rank-1 | mAP | Rank-1 | mAP | Rank-1 | mAP | |
| Market1501 --> DukeMTMC | 80.2 | 67.2 | 82.6 | 70.3 | 83.3 | 70.3 |
| DukeMTMC --> Market1501 | 91.7 | 79.3 | 93.6 | 82.2 | 94.2 | 83.0 |
| Market1501 --> MSMT17 | 51.6 | 26.6 | - | - | 61.7 | 34.6 |
| DukeMTMC --> MSMT17 | 59.0 | 32.0 | - | - | 66.9 | 38.3 |
Our code is based on MMT
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent