Jaded-Encoding-Thaumaturgy/vs-denoise

Properly support field-based handling

Opened this issue · 0 comments

Currently, every denoiser accepts interlaced input, but none of them actually handle it correctly.

To denoise these types of sources properly, we must:

  1. Separate the fields into their own clips (so one clip that features all the top fields, and one that features all the bottom fields)
  2. Perform processing on both as per usual (this will double the processing required, but inevitable here)
  3. Interleave and double weave them back together, then slice as necessary.

Theoretically, nothing is stopping us from implementing these for every denoiser besides figuring out a good way to write all that, so I will likely implement these eventually in at least the following denoisers:

  • MVTools
  • BM3D (wolfram plugins, will not do mawen's version to keep it more in line with how it was originally written)
  • nl_means
  • wnnm

But I'll likely look at other denoisers we have wrappers for as well.