DLVFX

DLVFX

DLVFX is part of the NVidia VFX SDK, and requires RTX GPUs.

akarin.DLVFX(clip clip, int op[, float scale=1, float strength=0, int output_depth=clip.format.bits_per_sample, int num_streams=1])

There are three operation modes:

• op=0: artefact reduction. int strength controls the strength, use 0 or 1.
• op=1: super resolution, scale>1 controls the scale factor. int strength controls the enhancement strength.
• op=2: denoising. float strength controls the strength. (Not working yet.)

Usage Notes:

• Only 32-bit floating point RGB and 8-bit integer RGB24 clips are supported as input clip.
• The output defaults to the same format as the input, however, you can set output_depth to 32 (RGBS) or 8 (RGB24) to override the default.
• Setting num_streams>1 will improve the performance by parallelizing processing of multiple frames on the GPU and will improve performance, as long as your GPU is capable enough to handle it.

This filter requires appropriate Video Effects library (v0.6 beta) to be installed.

Usage Hints

1. I've seen reports that op=0 might produce distorted colors. If that's a concern, you can always input a RGB gray clip by duplicating the Y channel to R/G/B channels. Most of the encoding artefacts are on the Y channel anyway.