I created an automatic script generator (read: by hand) so I could play with strength=0,1-1,0 and searchw=5/7/9 with all the possible permutations, keeping blend=0, chroma=true.
Unfortunately it doesn't work well as SMDegrain prefilter as it creates color/noise fluctuations in omogeneous areas.
I created an automatic script generator (read: by hand) so I could play with strength=0,1-1,0 and searchw=5/7/9 with all the possible permutations, keeping blend=0, chroma=true.
Unfortunately it doesn't work well as SMDegrain prefilter as it creates color/noise fluctuations in omogeneous areas.
I don't know much about SMDegrain but I can't see how DGDenoise could cause "color/noise fluctuations in homogeneous areas". Maybe with a script and link to SMDegrain I could have a look. If SMDegrain is going to require beaucoup extra stuff and a research project to get it working, however, maybe not.
I don't know much about SMDegrain but I can't see how DGDemoise could cause "color/noise fluctuations in homogeneous areas". Maybe with a script and link to SMDegrain I could have a look. If SMDegrain is going to require beaucoup extra stuff and a research project to get it working, however, maybe not.
I strongly recomend you to use Prefetch and test on really noisy stuff. I can send you some video comparison, if needed, as screenshot are useless in this case.
Filename is self explanatory: when you read DGDenoise is it only, when there is a "+" it is a combination with a prefilter and SMDegrain. When a prefilter is present, I already applied the best parameters, while on DGDenoise I scrolled thru all the possible strenght values.
Pay attention to some areas, such as on 00:06, the wall behind red chairs and play it step by step. Or other scenes too.
You can clearly see that KNL+SM is very clear, detailed and steady.
Not meaning to be rude to one of my good friends, but again, can you demonstrate a problem with DGDenoise alone? I'm not interested in the rest of the process. That's not on me. If you don't like it as a "prefilter" then don't use it. Not all tools are suitable for all purposes.
I will say that DGDenoise() is spatial only, whereas KNL is spatiotemporal. If there is temporal chroma noise that could be relevant.
As a general rule, should I use "strong" denoise parameters to use it as reference for motion compensated denoising?
Not sure whether there is a general rule. Your source seems to be low contrast so that motion vectors may not detect details. Perhaps try to feed a clear and high contrast image as a prefilter.
Your filtered samples seem to be overfiltered (flat) to my taste. You didn't include the unfiltered original in your samples though.
Can you please cut a sample? Don't want to download 5GB to look at a few frames. What is the source type, please? I will tell you how to cut a small sample. VOBs?
Done. I extracted some samples with AviDemux to preserve quality and pick the right sequence.
Look at the walls, the last mkv above all. You have a lot of noise and flashing.