linkserendipity/AP3D

Pytorch implementation of "Appearance-Preserving 3D Convolution for Video-based Person Re-identification"

Appearance-Preserving 3D Convolution for Video-based Person Re-identification

Requirements: Python=3.6 and Pytorch=1.0.0

Training and test

# For MARS
python train.py --root /home/guxinqian/data/ -d mars --arch ap3dres50 --gpu 0,1 --save_dir log-mars-ap3d #
python test-all.py --root /home/guxinqian/data/ -d mars --arch ap3dres50 --gpu 0 --resume log-mars-ap3d

Citation

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

@inproceedings{gu2020AP3D,
  title={Appearance-Preserving 3D Convolution for Video-based Person Re-identification},
  author={Gu, Xinqian and Chang, Hong and Ma, Bingpeng and Zhang, Hongkai and Chen, Xilin},
  booktitle={ECCV},
  year={2020},
}