How to Improve SLP Results

Topaz Video General
1

Suggestions for improving SLP based on long-term use and observation

After using SLP extensively over a long period, I’ve noticed a few areas where the model could be improved. Overall, the results are excellent, but there are a few recurring characteristics that, if addressed, would make SLP even stronger.

To summarize, SLP’s pixel structure is slightly too coarse. By this I mean that the color transitions between neighboring pixels are a bit too pronounced, making the image appear slightly “blocky.” The blocking is not broken up enough, which gives the output a somewhat harsh pixel structure.

The following improvements would significantly enhance the model, all 3 of them:

  • 1) Slight smoothing and denoising, similar to FFmpeg’s hqdn3d filter, value “hqdn3d=3:3:2:2” would fit for most. This is a 3D denoiser that operates across the X and Y spatial dimensions as well as the temporal dimension (consecutive frames). A very subtle amount would help soften the coarse pixel structure without sacrificing detail.
  • 2) A very light dehalo, comparable to the FineDehalo filter, at roughly a strength of 0.15 (depending on the source material). This would help reduce minor halo artifacts while preserving sharpness.
  • 3) A light compression artifact cleanup, similar to the “Fix Compression” feature in Iris MQ. A strength of around 15 seems to work well for a wide variety of sources and would help reduce residual blocking and compression artifacts.

Some capabilities already exist in other Topaz models, such as Iris and Proteus. I think it would be a great time to leverage these synergies and incorporate them into SLP. Doing so would preserve SLP’s excellent detail reconstruction while producing a cleaner, more natural, and more refined image.

3 Likes