In this repo, we include the runner-up solution of AICITY Challenge Track2 (Vehicle Re-Identification) at CVPR 2021 Workshop.

Robust Vehicle Re-identification via Rigid Structure Prior
Minyue Jiang*, Xuanmeng Zhang*, Yue Yu, Zechen Bai, Zhedong Zheng,Zhigang Wang, Jian Wang, Xiao Tan, Hao Sun, Errui Ding, Yi Yang
* denotes equal contribution.

The repo contains the training code, testing code and additional annotations on the training sets.

├── AICITY2021-Track2/
│   ├── annotations/
│   ├── global-aware-model/
│   ├── part-aware-model/
│   ├── post-processing/

1. Download the pre-trained model.
   • ResNet101-ibn
   • Se_ResNet101-ibn
   • ResNeSt-101
   • HRNet-48
   • Res2Net-200
2. Download the AICity2021 dataset.
3. Prepare the dataset. See prepare data .

We select images from the training set of CityFlowV2 to annotate the defined bounding box and the direction in annotations.

1. Follow reid model to train the reid models.
2. Follow detection model to train the detection models.
3. Follow direction model to train the direction classification models.

1. Follow reid model to extract the reid features.
2. Follow detection model to inference on query and test datasets.
3. Follow direction model to inference on query and test datasets.

Follow post-processing to get the final result.

AICITY2020 Challange Track2 Leaderboard

Our implementation is partly based on the following codebases. We gratefully thank the authors for their wonderful works: fast-reid, PaddleReid, and AICIty-reID-2020.

If you find our work useful in your research, please consider citing:

@inproceedings{jiang2021robust,
  author    = {Jiang, Minyue and Zhang, Xuanmeng and Yu, Yue and Bai, Zechen and Zheng, Zhedong and Wang, Zhigang and Wang, Jian and Tan, Xiao and Sun, Hao and Ding, Errui and Yang, Yi},
    title     = {Robust Vehicle Re-Identification via Rigid Structure Prior},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2021},
    pages     = {4026-4033}
}