HuPR: A Benchmark for Human Pose Estimation Using Millimeter Wave Radar

This is the official implementation of HuPR: A Benchmark for Human Pose Estimation Using Millimeter Wave Radar

Please cite our WACV 2023 paper if our paper/implementation is helpful for your research:

@InProceedings{Lee_2023_WACV,
    author    = {Lee, Shih-Po and Kini, Niraj Prakash and Peng, Wen-Hsiao and Ma, Ching-Wen and Hwang, Jenq-Neng},
    title     = {HuPR: A Benchmark for Human Pose Estimation Using Millimeter Wave Radar},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {January},
    year      = {2023},
    pages     = {5715-5724}
}

Preparation

Setup the conda environment.

conda env create -f environment.yml

Run setup.py to generate the directories needed

Dowload the dataset and annotations from the following link HuPR dataset

Please request access with your institutional email id and please provide following information

• Your Full Name
• Institution
• Advisor/Supervisor Name
• Current Position/Title
• Emaill Address (with institutional domain name)
• Purpose

Extract the dataset in the 'preprocessing/raw_data/iwr1843'

Preprocessing

Preprocess the raw radar data collected by two radar sensors (IWR1843Boost)

cd preprocessing
python process_iwr1843

Each dataset structure should be aligned in this way:

data/HuPR
    - hrnet_annot_test.json
    - hrnet_annot_val.json
    - hrnet_annot_train.json
    - single_1
        - hori
            - 000000000.npy
            .
            .
            .
        - vert
            - 000000000.npy
            .
            .
            .
        - visualization
    - single_2
    .
    .
    .

Specify the root of dataset HuPR in YOUR_CONFIG.yaml -> DATASET -> dataDir

Download the PythonAPI of COCO here

Replace coco.py and cocoeval.py with misc/coco.py and misc/cocoeval.py respectively

Training

python main.py --config <YOUR_CONFIG>.yaml --dir <YOUR_DIR>

YOUR_DIR specifies the diretory that will be saved in logs/

e.g.

python main.py --config mscsa_prgcn.yaml --dir mscsa_prgcn

To facilitate the training speed, you may set -sr 10 to the training command

Evaluation

python main.py --dir <YOUR_DIR> --config <YOUR_CONFIG>.yaml --eval

YOUR_DIR specifies the diretory that will be loaded from logs/

The loaded weighted should be named as 'model_best.pth'

Download the trained weight from the drive

Evaluate model performance and visualize the results

python main.py --dir <YOUR_DIR> --config <YOUR_CONFIG>.yaml --eval --vis <YOUR_VIS_DIR>

YOUR_VIS_DIR specifies the directory where the results will be saved in visualization/

e.g.

python main.py --config mscsa_prgcn.yaml --dir mscsa_prgcn --eval