SimHMR is a simple and effective framework for 3D human mesh recovery from single images. It combines the power of transformer architectures with SMPL body model to achieve state-of-the-art performance on various benchmarks.

• Python 3.8
• CUDA 10.2+ (for GPU training)
• Conda (recommended for environment management)

1. Clone the repository:

git clone https://github.com/Inso-13/simhmr.git
cd simhmr

2. Create and activate conda environment:

conda env create -f env.yml
conda activate human

3. Install SimHMR:

pip install -e .

1. Clone the repository:

git clone https://github.com//Inso-13/simhmr.git
cd simhmr

2. Install PyTorch (CUDA 10.2):

pip install torch==1.8.0+cu111 torchvision==0.9.0+cu111 torchaudio==0.8.0 -f https://download.pytorch.org/whl/torch_stable.html

3. Install other dependencies:

pip install -r requirements.txt

4. Install SimHMR:

pip install -e .

• PyTorch: 1.8.0
• MMCV: 1.5.3
• MMDetection: 2.27.0
• MMPose: 0.28.1
• SMPL-X: 0.1.28
• PyTorch3D: 0.7.2
• OpenCV: 4.7.0.68

1. Prepare your dataset and update the configuration file
2. Start training:

python tools/train.py configs/simhmr/pw3d.py

python tools/test.py configs/simhmr/pw3d.py work_dirs/checkpoint.pth --eval mpjpe pa-mpjpe

Please organise all datasets under the data/ directory following the MMHuman3D format. See MMHuman3D documentation for details.

If you find this work useful, please cite:

@inproceedings{huang2023simhmr,
  title={Simhmr: A simple query-based framework for parameterized human mesh reconstruction},
  author={Huang, Zihao and Shi, Min and Liu, Chengxin and Xian, Ke and Cao, Zhiguo},
  booktitle={Proceedings of the 31st ACM International Conference on Multimedia},
  pages={6918--6927},
  year={2023}
}

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

• This work builds upon MMHuman3D
• Thanks to the SMPL model authors
• Thanks to all the dataset providers

We welcome contributions! Please see our contributing guidelines for details.