Temporal Knowledge Propagation for Image-to-Video Person Re-identification

Requirements: Python=3.6 and Pytorch=1.0.0

Training and test

# For MARS
python train.py --root /data/datasets/ -d mars --save_dir log-mars
python test.py --root /data/datasets/ -d mars --resume log-mars/best_model.pth.tar --save_dir log-mars

# For DukeMTMC-VideoReID
python train.py --root /data/datasets/ -d dukevid --save_dir log-duke
python test.py --root /data/datasets/ -d dukevid --resume log-duke/best_model.pth.tar --save_dir log-duke

# For iLIDS-VID (If you use the pretrained model on Duke, you will get a much higher results than that reported in our paper.)
python main_ilids.py --root /data/datasets/ --save_dir log-ilids

Citation

If you use our code in your research or wish to refer to the baseline results, please use the following BibTeX entry.

@inproceedings{gu2019TKP,
  title={Temporal Knowledge Propagation for Image-to-Video Person Re-identification},
  author={Gu, Xinqian and Ma, Bingpeng and Chang, Hong and Shan, Shiguang and Chen, Xilin},
  booktitle={ICCV},
  year={2019},
}