README.md

Pedestrian Attribute Recognition

📚 Contents

• Pedestrian Attribute Recognition
  • 📚 Contents
  • 📋 Introduction
  • 📂 Datasets
  • 🛠️ Environment
  • 🚀 Get Started
  • 💗 Acknowledgement
  • 🤝 Contribute & Contact

📋 Introduction

The HAP pre-trained model is fine-tuned for the pedestrian attribute recognition task with respect to:

• Three datasets: PA-100K, RAP, and PETA

📂 Datasets

Put the dataset directories outside the HAP project:

home
├── HAP
├── PA-100K  # PA-100K dataset directory
│   ├── annotation.mat
│   ├── dataset_all.pkl
│   └── data
│       ├── xxx.jpg
│       └── ...
├── RAP  # RAP dataset directory
│   ├── RAP_annotation
│   │   └── RAP_annotation.mat
│   ├── dataset_all.pkl
│   └── RAP_dataset
│       ├── xxx.png
│       └── ...     
└── PETA  # PETA dataset directory
    ├── PETA.mat
    ├── dataset_all.pkl
    └── images
        ├── xxx.png
        └── ...

🛠️ Environment

Conda is recommended for configuring the environment:

conda env create -f env_attribute.yaml && conda activate env_attribute

# Install mmcv
cd ../2d_pose_estimation/mmcv && git checkout v1.3.9 && MMCV_WITH_OPS=1 && cd .. && python -m pip install -e mmcv

🚀 Get Started

It may need 8 GPUs with memory larger than 6GB, such as NVIDIA V100, for training.

# -------------------- Fine-Tuning HAP for Pedestrian Attribute Recognition --------------------

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

cd Rethinking_of_PAR/

DATA=pa100k  # {pa100k, rapv1, peta}

python -m torch.distributed.launch \
    --nproc_per_node=${NPROC_PER_NODE} \
    --master_port=${PORT} \
    train.py \
    --cfg configs/pedes_baseline/${DATA}.yaml

💗 Acknowledgement

Our implementation is based on the codebase of Rethinking_of_PAR .

🤝 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]).