This repo contains the codebase of our TPAMI 2020 paper: Learning Multi-Attention Context Graph for Group-Based Re-Identification.
We use python 3.7 and pytorch=1.0.1 torchvision=0.2.1
All experiments are done on CSG. Please download CUHK-SYSU.
Original dataset webpage: CUHK-SYSU Person Search Dataset or JDE dataset zoo
The group annotation files are in ./data
To train the model, please run
python main_group_gcn_siamese_part_half_fulltest_sink.py --data-root [data path] --max-epoch 300 --stepsize 100 --eval-step 300 --gpu-devices 0
An example can be found in run_train_test.sh
To evaluate the trained model, run
python main_group_gcn_siamese_part_half_fulltest_sink.py --data-root [data path] --evaluate True --gpu-devices 0 --pretrained-model [model path]
| Task | Rank1 | mAP |
|---|---|---|
| Group re-id | 65.5% | 67.0% |
| Person re-id | 65.1% | 64.2% |
Link of the trained model: [Google]
Our code is developed based on Video-Person-ReID (https://github.com/jiyanggao/Video-Person-ReID).
If you find this repo useful in your project or research, please consider citing it:
@ARTICLE{9233968,
author={Y. {Yan} and J. {Qin} and B. {Ni} and J. {Chen} and L. {Liu} and F. {Zhu} and W. -S. {Zheng} and X. {Yang} and L. {Shao}},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
title={Learning Multi-Attention Context Graph for Group-Based Re-Identification},
year={2020},
doi={10.1109/TPAMI.2020.3032542}}
}