Skip to content

Latest commit

 

History

History
96 lines (78 loc) · 2.53 KB

README.md

File metadata and controls

96 lines (78 loc) · 2.53 KB
Raw

Person ReID

📚 Contents

📋 Introduction

The HAP pre-trained model is fine-tuned for the conventional person ReID task with respect to:

  • Two datasets: MSMT17 and Market-1501
  • Two resolution sizes: (256, 128) and (384, 128)
  • Two model structures: ViT and ViT-lem

📂 Datasets

Put the dataset directories outside the HAP project:

home
├── HAP
├── msmt  # MSMT17 dataset directory
│   ├── bounding_box_train
│   │   ├── xxx.jpg
│   │   └── ...
│   ├── bounding_box_test
│   │   ├── xxx.jpg
│   │   └── ...
│   └── query
│       ├── xxx.jpg
│       └── ...
└── market  # Market-1501 dataset directory
    ├── bounding_box_train
    │   ├── xxx.jpg
    │   └── ...
    ├── bounding_box_test
    │   ├── xxx.jpg
    │   └── ...
    └── query
        ├── xxx.jpg
        └── ...

🛠️ Environment

Conda is recommended for configuring the environment:

conda env create -f env-person_reid.yaml && conda activate env_person_reid

🚀 Get Started

It may need 1 GPU with memory larger than 12GB, such as NVIDIA V100, for training.

# -------------------- Fine-Tuning HAP for Person ReID --------------------
cd HAP/downstream/person_reid/

# Download the checkpoint and move it here
CKPT=ckpt_default_pretrain_pose_mae_vit_base_patch16_LUPersonPose_399.pth

GPU=0
SEED=0
TAG=default
SIZE=256  # {256, 384}
DATA=msmt  # {msmt, market}
MODEL=vit_base_patch16  # {vit_base_patch16, lem_base_patch16}
OUTPUT=output-person_reid/${TAG}/${DATA}/${MODEL}/${SIZE}/${SEED}/

python main_reid.py \
    --config_file configs/reid/${DATA}.yaml \
    --model ${MODEL} \
    --batch_size 64 \
    --resume ${CKPT} \
    --epochs 100 \
    --warmup_epochs 5 \
    --lr 8e-3 \
    --size ${SIZE} \
    --device ${GPU} \
    --seed ${SEED} \
    --output_dir ${OUTPUT}

💗 Acknowledgement

Our implementation is based on the codebase of MALE.

🤝 Contribute & Contact

Feel free to star and contribute to our repository.

If you have any questions or advice, contact us through GitHub issues or email ([email protected]).