I love NVEncC and it's growing array of filters, but one thing I think that's lacking is a decent temporal noise reduction filter with sufficient controls. I've tested the FFT3D filter, and while it certainly seems better than the convolution3D filter, I've noticed that there could be a few more granular controls exposed, as detailed here: http://avisynth.nl/index.php/FFT3DGPU
The key ones for me are:
- bt - in the current nvencc version we have a temporal=0/1 option, which I assume then uses the previous and current frame, but this gives further options up to 5 frames (2 previous, current & 2 next) which can give superior results.
- sigma2/3/4 - to set amount of denoise for med/large/huge noise frequencies.
In some cases I've had to resort to running videos through NeatVideo first, which is the best temporal denoiser, but it annoyingly doesn't have a cli interface so is a lot more work having to run through an NLE first. Adding the above params for fft3d should get us a lot closer to that natively.
I love NVEncC and it's growing array of filters, but one thing I think that's lacking is a decent temporal noise reduction filter with sufficient controls. I've tested the FFT3D filter, and while it certainly seems better than the convolution3D filter, I've noticed that there could be a few more granular controls exposed, as detailed here: http://avisynth.nl/index.php/FFT3DGPU
The key ones for me are:
In some cases I've had to resort to running videos through NeatVideo first, which is the best temporal denoiser, but it annoyingly doesn't have a cli interface so is a lot more work having to run through an NLE first. Adding the above params for fft3d should get us a lot closer to that natively.