Sharpened Cosine Distance
PyTorch implementation of the Sharpened Cosine Distance operator.
The core idea came from Brandon Rohrer(@brohrer) and the implementation is based on the tf/keras implementation of Raphael Pisoni.
This implementation supports
- 2D operation only
- asymmetric kernels, any shape
- CUDA / GPU
If you find this implementation useful please give it a star. Open issues for bugs/ideas.
If you are planning to build something on top of it let me know, I am always up for some good collaborations ;)