Filter suggestions (using non-ml filters)?

[Selur]

For just fun and curiosity, I'm trying to filter the attached video (from NTIRE; also attached) with non-machine learning based filters (ideally as automated as possible).

And I was wondering how others would tackle this clip.
My current approach is using:

clip = smdegrain.SMDegrain(clip, tr=3, thSAD=1500, thSADC=256, RefineMotion=True, contrasharp=50, interlaced=False, subpixel=3, prefilter=3, truemotion=True, MVglobal=True, dct=1, chroma=False, opencl=True, device=-1)

which is horribly slow, due to the dct=1. (attached result as SMDegrain.mp4)
This works fine for the first scene, but has problems with the 2nd scene. (also the text probably should be filtered with some mask)

So, how would you tackle this with conventional filters?

Cu Selur

Ps.: RefineMotion=False works a bit better

[Sharc]

For just fun and curiosity, I'm trying to filter the attached video (from NTIRE; also attached) with non-machine learning based filters (ideally as automated as possible).

And I was wondering how others would tackle this clip.
My current approach is using:

clip = smdegrain.SMDegrain(clip, tr=3, thSAD=1500, thSADC=256, RefineMotion=True, contrasharp=50, interlaced=False, subpixel=3, prefilter=3, truemotion=True, MVglobal=True, dct=1, chroma=False, opencl=True, device=-1)

which is horribly slow, due to the dct=1. (attached result as SMDegrain.mp4)
This works fine for the first scene, but has problems with the 2nd scene. (also the text probably should be filtered with some mask)

So, how would you tackle this with conventional filters?

Cu Selur

Ps.: RefineMotion=False works a bit better

A tough case for am automatic fix, it seems ....

[Selur]

Yes, otherwise it wouldn't be fun.
When handling such a thing manually (given that the source it worth the effort) I would probably adjust the denoising between scenes, apply different filtering around the subtitles and similar stuff.

Cu Selur