Neat video alternative in vapoursynth

[vigan1]

Hello. Is there anything close or better than neatvideo in vapoursynth ?

Did anyone made a benchmark with free denoiser and neat video ?
Like this :
https://deepai.org/publication/opendenoising-an-extensible-benchmark-for-building-comparative-studies-of-image-denoisers

With image like those :
https://images.deepai.org/converted-papers/1910.08328/x4.png
https://www.semanticscholar.org/paper/A-High-Quality-Denoising-Dataset-for-Smartphone-Abdelhamed-Lin/ebf35073e122782f685a0d6c231622412f28a53b/figure/1
Thank you.

[lansing]

There is none.

[feisty2]

There’s a V-BM3D plugin for vaporsynth

[Cary Knoop]

There’s a V-BM3D plugin for vaporsynth
That's a great plugin that unfortunately is not available in a GPU format.

Neat video is way overrated.

Also highly compressed video does not have much noise in the general sense of the word.

[feisty2]

you can increase block_step to make BM3D significantly faster, at the cost of lower processing precision.

[vigan1]

Thank you guys. So there is no whitepaper style comparison between free denoiser algorithms and neat video. Too bad.

I feel neat video is overrated too (and overpriced), but it's fast and clean.

But I wasn't talking about processing speed, I want to know if you guys have a method that provides better results than neatvideo consistently. I mainly used neatVideo to break banding in 8bit420 codecs, and clean the remaining noise. Do you know the best denoiser algorithm and settings to do the job as well as NV ?

Maybe NLmeans (is GPU accelerated in vapoursyth, I think ?) would be the closest in term of speed and quality ?

BM3D in the FAST preset, seems to become my Denoiser of choice. But it's hard to understand every parameter, so if you can provide workflow or settings or vpy, It would be perfect.

Thank you.

[feisty2]

BM3D is one of the highest quality non-learning based denoising algorithms, neural net based models may easily surpass the denoising quality of BM3D in some cases but may fail miserably in other cases, its not that stable like traditionally methods so its hard to say if it could consistently outperform BM3D

[vigan1]

Thank you feisty2.
Using it a little, I felt that V-BM3D is a beast of a denoiser, that is why I wanted to stick with it, sometime it produce results that are unexpectedly clean and extremely sharp.

Do you know on what algorithm is neat video based on ?
It's probably not as good as BM3D or NLmeans.