Home

Awesome

LearnableGroups-Hand

The code for the paper Exploiting Learnable Joint Groups for Hand Pose Estimation (Accepted by AAAI2021).

Paper

Overall network:

<p align="middle"> <img src="assets/overall_network.png", width="780"> </p>

Qualitative Results

some qualitative results on the RHD/STB/FHD dtasets. In each triplet, from left to right: imgs (input), predictions, GT.

<p align="middle"> <img src="assets/RHD_qualitative_1.png", width="780"> </p> <p align="middle"> <img src="assets/FHD_qualitative_1.png", width="780"> </p> <p align="middle"> <img src="assets/STB_qualitative_1.png", width="780"> </p>

Citing LearnableGroups-Hand

If this repository is helpful to your research, please cite the paper:

@misc{li2020exploiting,
      title={Exploiting Learnable Joint Groups for Hand Pose Estimation}, 
      author={Moran Li and Yuan Gao and Nong Sang},
      year={2020},
      eprint={2012.09496},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Usage

The code is built on Python3 and Pytorch 1.6.0.

Install dependencies

pip install -r requirements.txt

Run the code

python eval_RHD.py --data_dir 'your RHD_published_v2 dataset path'

Comparison with SOTA methods