Mr-Z-2697

Results 3 comments of Mr-Z-2697

> The current implementation is specialized for `group_size=8`, otherwise the speed may drop by at most 2x. I wish you could add this parameter. Your implementation is amazing, the speed...

> Have you tried `WNNM`? It is a high quality denoiser but with tunable `group_size` I've tried it. In fact, this thought initially came from my testing on WNNM, then...

> ~Which backend do you prefer, `CPU` or `CUDA`? It is much more easy to implement the `CPU` backend based on `WNNM` and FFTW.~ > > Considering the efforts required,...