TransMatcher: Deep Image Matching Through Transformers for Generalizable Person Re-identification
This is the official PyTorch code for the TransMatcher proposed in our paper [1].
For further details, please read our paper, and a poster here.
It is based on the QAConv 2.0 code, and the requirements and usage are quite similar. For a quick run, please try the demo.sh. Ignore the accuracy of this demo, since it is only for validating that everything is OK to run.
Performance (%) of TransMatcher under direct cross-dataset evaluation without transfer learning or domain adaptation:
| Training Data | Method | CUHK03-NP | Market-1501 | MSMT17 | |||
| Rank-1 | mAP | Rank-1 | mAP | Rank-1 | mAP | ||
| Market | QAConv 2.0 | 16.4 | 15.7 | - | - | 41.2 | 15.0 |
| TransMatcher | 22.2 | 21.4 | - | - | 47.3 | 18.4 | |
| MSMT | QAConv 2.0 | 20.0 | 19.2 | 75.1 | 46.7 | - | - |
| TransMatcher | 23.7 | 22.5 | 80.1 | 52.0 | - | - | |
| MSMT (all) | QAConv 2.0 | 27.2 | 27.1 | 80.6 | 55.6 | - | - |
| TransMatcher | 31.9 | 30.7 | 82.6 | 58.4 | - | - | |
| RandPerson | QAConv 2.0 | 14.8 | 13.4 | 74.0 | 43.8 | 42.4 | 14.4 |
| TransMatcher | 17.1 | 16.0 | 77.3 | 49.1 | 48.3 | 17.7 | |
Shengcai Liao
Inception Institute of Artificial Intelligence (IIAI)
[email protected]
[1] Shengcai Liao and Ling Shao, "TransMatcher: Deep Image Matching Through Transformers for Generalizable Person Re-identification." In Neural Information Processing Systems (NeurIPS), 2021.
@article{Liao-NeurIPS2021-TransMatcher,
author = {Shengcai Liao and Ling Shao},
title = {{TransMatcher: Deep Image Matching Through Transformers for Generalizable Person Re-identification}},
booktitle = {Neural Information Processing Systems (NeurIPS)},
year={2021}
}