Daily builds version 2

[ronaldjung]

I think the mistakes in this test are:
1) Using pass limit instead of time limit - this does not tell us anything. Maybe the rendering was super fast in one case, and slow in the other, or maybe not...? So I would suggest using time limit.
2) Using unrealistic number of passes rendered and resolution. There was simply not enough passes to render details in reasonable quality, denoised or not. The resolution is also so low that we cannot really see the details, and this is not something you would do in real work.

I didn´t have time to do this test, just tried to do this quickly to see the 5% improvement. But still very glad with new results! I know I didnt made a correct metodology to the test and even thought not to post. Just did to share. Maybe to the developers know that, at least in low resolution and time limit, the noise level increased. Also, I used the New Solver, maybe this have something to do with the result.
Maybe someone with more time available can make this tests with more liability.

yes, IMHO it is now looking better, because the denoiser is preserving the details of the map. I am hoping this will allow people to use denoising in more situations

I saw that! Is much more better now.
And recently I'm trying to use less denoise in my renders, the New Solver almost made it obsolete, very quick! Always usins it.

Regards

Ronald

[TomG]

Bear in mind that the "noise level" is not an exact science at the end of the day, and your eyes should be the final judge :) As part of moving to 1.7, for example, you could raise the noise level from 5% to 6% or even 7% and get visually the same amount of "cleanliness" in the image. There could be  something there worth investigating, but ultimately if a noise level result changes from 6% to 6.5% but the image looks better and renders just as fast or faster, then your eyes are the true judge and the shift in measured noise level is just that, a shift in how it is measured and not a shift in actual image quality.

[romullus]

Just stumbled on the case where simple mesh with very basic material with blurry reflections, lit by only HDRI, shows distinctive patches of noise, which takes unreasonable time to resolve. Switching from new improved sampler to stratified, completely cures this problem. I vaguely remember something like that being reported. Is this acknowledged or do you want simple scene with reproducible bug? Build Dec 14.

[maru]

Just stumbled on the case where simple mesh with very basic material with blurry reflections, lit by only HDRI, shows distinctive patches of noise, which takes unreasonable time to resolve. Switching from new improved sampler to stratified, completely cures this problem. I vaguely remember something like that being reported. Is this acknowledged or do you want simple scene with reproducible bug? Build Dec 14.

Yes please. Scene will be appreciated.
The only noisy patches I know of are caused by the new light solver, not by the DMC sampler...

[romullus]

Hm, that is strange, i found that it's also somehow related to CoronaBitmap's dome mode. Maybe it's faulty HDRI, but i used that HDRI without issues countless times. Anyway, i packed HDRI together with scene file. Also attaching screens. Notice how it's artifacts free only with stratified sampler AND spherical mode. Other combinations produces nasty artifacts.

Uploaded file: 1513378419_Improved-Sampler-Noise.7z

[pokoy]

That's exactly the same issue I came across and reported before 1.6, I'll have a look if I can find the original scene and report.

[maru]

Have you tried changing "enviro sampler" to something else?

[romullus]

Not before you asked. It seems like every enviro sampler except normal dependent, solves both issues. But fast compensated cleans noise very slowly if combined with CoronaBitmaps spherical mode. In dome mode all enviro samplers cleans noise noticeably faster than in spherical mode... Damn, it's so confusing, i just wish that Corona would render great again with default settings, like it used to do :/

[romullus]

BTW, i just found in my HDD earlier rendered image, which shows exactly the same artifacts like in 03 and 04 screens from post #64. That image was rendered in May 30 most likely with latest daily build of that date. And exactly the same HDRI was used... in dome mode. So it's probably not new improved random sampler at fault.

[sprayer]

romullus i have similar issues, i think this is something with reflection of hdri and geometry. for example i was made simple spline shape lathe it and it gives similar noise at core, and nothing can do weld core, weld vertixes, smooth, xform, nothing removing this noise in center. But at new scene it can't be reproduces. This hard to see at complex scene but at product renders... Scene i was send to Ondra couples days ago

[romullus]

In my case geometry is very simple, just a plane primitive with bend and turbosmooth modifiers on top, surely that can't cause any artifacts. It maybe that this particular HDRI is corrupted, i'll try to reproduce the bug with another one.

EDIT: just testesd - i get those artifacts with any HDRI i've tried.

[maru]

Romullus, have you already reported this somewhere else than in this forum thread? Mantis? It sounds like something that should be fixed urgently.

[romullus]

Nope, only here. I have uploaded scene on private uploader with reference to this discussion though. Wouldn't it be enough?

[maru]

Nope, only here. I have uploaded scene on private uploader with reference to this discussion though. Wouldn't it be enough?

It can easily get forgotten. It would be best if you could write a report on mantis and mention the uploaded file name there, or at least start a separate topic on the forum.

[romullus]

Ok, will do.