This source code is the conversion of a demo source code implemented in C++ into a pytorch.
You can download hand.pts
in this link
Any contributions or pull requests would be welcome!
Unfortunately, the original code is not suitable for actual use. So I added more improvements compared to the original code.
- This code uses a cyclic detection method mentioned in the paper. Therefore, even small hand key points can be detected with high accuracy.
- The original code inefficiently uses the O(N^4) method to find the local maximum value. I use
skimage
library to optimize this task. and this approach shows more than 40x performance in speed! (+25fps in GTX970)
- skimage
- pytorch (w/ CUDA)
- hand.pts
- opencv2
- python 3.7+
- SRHandNet website
- paper link
- original code
- Yangang Wang, Baowen Zhang and Cong Peng. "SRHandNet: Real-time 2D Hand Pose Estimation with Simultaneous Region Localization". IEEE Transactions on Image Processing, 29(1):2977 - 2986, 2020.