/PersonReID-VAAL

Code for AAAI 2020 paper Viewpoint-Aware Loss with Angular Regularization for Person Re-Identification.

Primary LanguagePythonOtherNOASSERTION

Viewpoint-Aware Loss with Angular Regularization for Person Re-Identification (AAAI 2020)

Code for AAAI 2020 paper Viewpoint-Aware Loss with Angular Regularization for Person Re-Identification.

If you find this code useful in your research, please consider citing:

@article{zhu2020viewpoint,
  title={Viewpoint-Aware Loss with Angular Regularization for Person Re-Identification},
  author={Zhihui Zhu, Xinyang Jiang, Feng Zheng, Xiaowei Guo, Feiyue Huang, Weishi Zheng, Xing Sun},
  booktitle={AAAI},
  year={2020}
}

Requirements

pytorch>=0.4 torchvision ignite=0.1.2 (Note: V0.2.0 may result in an error) yacs

Training

  1. Download the public datasets ( market1501 and DukeMTMC are supported) and use the corresponding dataloader.
  2. Construct viewpoint information. For each image, assign four labels: front, back, right side and left side The viewpoint meta data forms a dictionary and stores in a pickle file: {'image_name.jpg':(0/1/2/3, )}
  3. To use your own dataset re-implement the dataloader in directory "data/datasets".
  4. Sample running command under the same directory of this readme file: training config is stored in a yaml file, the examples are in configs directory sh market_run.sh / sh duke_run.sh

Model Framework

Framework

Model Performance

Performance0 Performance0