LuxCoreRender Forums

What about RPR denoiser ? this engine is opensource.

https://www.youtube.com/watch?v=4ETxx9dff4A

Looks awesome (theoretically)

Well, I have no experience with RPR but that definitely doesn't look impressive at all... But maybe on a more complex example with some textures and dof and so on it could be good I don't know. The only open source solution I've seen doing good was cycles's denoiser.

cycles's denoiser.

Do you have an idea of the paper behind Cycles denoiser .?

Sharlybg wrote: ↑Fri Nov 02, 2018 2:46 pm

cycles's denoiser.

Do you have an idea of the paper behind Cycles denoiser .?

Absolutely not, I have also never written a line of code and suck at math .... Ok I've written some basic scripts but that's it.

After some googling for cycles denoiser fall into this :

http://sglab.kaist.ac.kr/WLR/

https://docs.opencv.org/trunk/d5/d69/tu ... #gsc.tab=0

https://sinmantyx.wordpress.com/2015/12 ... er-theory/

By the way the current denoiser is really enough for some kind of project. It help me a lot here :

viewtopic.php?f=3&t=567

Sharlybg wrote: ↑Fri Nov 02, 2018 3:10 pm By the way the current denoiser is really enough for some kind of project. It help me a lot here :

viewtopic.php?f=3&t=567

I said. I will stop complaining, so to this I will only say, good for you
(but I choose 500 samples+ ai denosier over 4000 samples+ BCD any day)

For animations here, done in Feb. 2017, first Blender experimental 2.77 with Cycles denoiser was used - 80 samples @ 1080p.
And now, patch D3889 which "... extends the denoiser to support using multiple input frames in order to denoise one frame & can significantly reduce noise and flickering in animations", awaits review.

BTW, I too have a feeling something is slightly off with present implementation of BCD.

kintuX wrote: ↑Sat Nov 03, 2018 1:18 am For animations here, done in Feb. 2017, first Blender experimental 2.77 with Cycles denoiser was used - 80 samples @ 1080p.
And now, patch D3889 which "... extends the denoiser to support using multiple input frames in order to denoise one frame & can significantly reduce noise and flickering in animations", awaits review.

BTW, I too have a feeling something is slightly off with present implementation of BCD.

It is not implementation it is by design. Best results you ever find are always about fine/residual noise removal. It is not designed for previews or quick one-off renderings with mediocre quality. It needs clear definition in rendering, wether it is textures or object details or it will fall apart. I was too for a long time thinking it is either broken, or needs to be tweaked to work better in luxcore. But no, it is simply not designed to work with low samples. (note that this is not a complaint )