[ACM MM 2020 Oral] Black Re-ID: A Head-shoulder Descriptor for the Challenging Problem of Person Re-Identification paper

2020-08-12: Update Code.

2020-08-20: Update paper link.

If you find the code useful, please consider citing our paper:

@InProceedings{xu2020ACM,
author = {Boqiang, Xu and Lingxiao, He and Xingyu, Liao and Wu,Liu and Zhenan, Sun and Tao, Mei},
title = {Black Re-ID: A Head-shoulder Descriptor for the Challenging Problem of Person Re-Identification},
booktitle = {Proceedings of the 28th ACM International Conference on Multimedia (MM '20)},
month = {October},
year = {2020}
}

• Dataset: Black re-ID (BaiDuDisk pwd:xubq please add the path of the Black re-ID dataset to DATASETS.DATASETS_ROOT in ./projects/Black_reid/configs/Base-HAA.yml)
• Pre-trained STN Model (BaiDuDisk pwd:xubq please add the path of the STN model to DATASETS.STN_ROOT in ./projects/Black_reid/configs/Base-HAA.yml)

1. cd to folder:

 cd projects/Black_reid

2. If you want to train with 1-GPU, run:

CUDA_VISIBLE_DEVICES=0 train_net.py --config-file= "configs/HAA_baseline_blackreid.yml"

if you want to train with 4-GPU, run:

CUDA_VISIBLE_DEVICES=0,1,2,3 train_net.py --config-file= "configs/HAA_baseline_blackreid.yml"

To evaluate a model's performance, use:

CUDA_VISIBLE_DEVICES=0 train_net.py --config-file= "configs/HAA_baseline_blackreid.yml" --eval-only MODEL.WEIGHTS /path/to/checkpoint_file

If you have any question about the project, please feel free to contact me.

E-mail: [email protected]

The code was developed based on the ’fast-reid’ toolbox https://github.com/JDAI-CV/fast-reid.