Skip to content

Latest commit

 

History

History
103 lines (70 loc) · 3.79 KB

README.md

File metadata and controls

103 lines (70 loc) · 3.79 KB
Raw

Python >=3.5 PyTorch >=1.6

Template-Aware Transformer for Person Reidentification

The official repository for Template-Aware Transformer for Person Reidentification.

Pipeline

framework

Requirements

Installation

pip install -r requirements.txt

We use / torch 1.7.1 / torchvision 0.8.2 / timm 0.4.12 / cuda 11.1 / 24GB RTX 3090 for training and evaluation.

Note that we use torch.cuda.amp to accelerate speed of training which requires pytorch >= 1.6.

Prepare Datasets

Download the person datasets Market-1501, MSMT17, DukeMTMC-reID, Occluded-Duke.

Then unzip them and rename them under the directory like:

data
├── dukemtmcreid
│   └── images ..
├── market1501
│   └── images ..
├── MSMT17
│   └── images ..
└── Occluded_Duke
    └── images ..

Prepare ViT Pre-trained Models

You need to download the ImageNet pretrained transformer model: ViT-Base.

Training

We utilize 4 GPUs for training. You can directly train with following yml and commands:

# Market
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 train.py --config_file configs/Market/vit_tat_stride_384.yml MODEL.DIST_TRAIN True SOLVER.BASE_LR 0.032

# DukeMTMC
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 train.py --config_file configs/DukeMTMC/vit_tat_stride_384.yml MODEL.DIST_TRAIN True SOLVER.BASE_LR 0.032

# OCC_Duke
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 train.py --config_file configs/OCC_Duke/vit_tat_stride.yml MODEL.DIST_TRAIN True SOLVER.BASE_LR 0.032

# MSMT17
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 train.py --config_file configs/MSMT17/vit_tat_stride_384.yml MODEL.DIST_TRAIN True SOLVER.BASE_LR 0.032

Evaluation

CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 test.py --config_file <file-path> TEST.WEIGHT <weight-path>

Some examples:

# Market
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 test.py --config_file configs/Market/vit_tat_stride_384.yml TEST.WEIGHT '../logs/market_vit_tat_stride_384/transformer_120.pth'

# DukeMTMC
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 test.py --config_file configs/DukeMTMC/vit_tat_stride_384.yml TEST.WEIGHT '../logs/duke_vit_tat_stride_384/transformer_120.pth'

# OCC_Duke
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 test.py --config_file configs/OCC_Duke/vit_tat_stride.yml TEST.WEIGHT '../logs/occ_duke_vit_tat_stride/transformer_120.pth'

# MSMT17
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 test.py --config_file configs/MSMT17/vit_tat_stride_384.yml TEST.WEIGHT '../logs/msmt17_vit_tat_stride_384/transformer_120.pth'

Acknowledgement

Codebase from TransReID.

Citation

Please cite this paper if it helps your research:

@article{zheng2022template,
  title={Template-Aware Transformer for Person Reidentification},
  author={Zheng, Yanwei and Zhao, Zengrui and Yu, Xiaowei and Yu, Dongxiao},
  journal={Computational Intelligence and Neuroscience},
  volume={2022},
  year={2022},
  publisher={Hindawi}
}