OpenImageDenoise

[B.Y.O.B.]

It may be a side effect of some internal Oidn behavior: Alpistinho is initializing the Oidn device and filter at the plugin allocation, we could do that for every apply() however it may be slower so I tend to agree with Alpistinho's decision.

If you are talking about the stuff in the constructor, this takes 0.2 ms on my notebook CPU:

Code: Select all

OIDN constructor took 0.000236034 seconds

So I would just move it to Apply.

[Dade]

So I would just move it to Apply.

Done.

[B.Y.O.B.]

Thanks, the memory hogging is gone.

[FarbigeWelt]

Second try, hopeful fixed now:

COMP-3x3-fixed.png

Does anybody else observe the lack of depth of denoised images in the first two rows compared to the noisy one in the same line. Is this an illusion of mine, this good visible distance between the left top edge of the plate and the wall in the background?

Beside that, it’s amazing how good oidn works. And even more, how fast and far you got!

[Dade]

Second try, hopeful fixed now:

COMP-3x3-fixed.png

Does anybody else observe the lack of depth of denoised images in the first two rows compared to the noisy one in the same line. Is this an illusion of mine, this good visible distance between the left top edge of the plate and the wall in the background?

Beside that, it’s amazing how good oidn works. And even more, how fast and far you got!

Left side of the image is brighter the the right side. When right side is attached to the left side of another rendering it may lead to perceive, a not existing, difference ... but it is really hard to judge as all human perception related things.

[lacilaci]

Second try, hopeful fixed now:

COMP-3x3-fixed.png

Does anybody else observe the lack of depth of denoised images in the first two rows compared to the noisy one in the same line. Is this an illusion of mine, this good visible distance between the left top edge of the plate and the wall in the background?

Beside that, it’s amazing how good oidn works. And even more, how fast and far you got!

The difference between noisy and denoised image is much much bigger to perception than people think.
Noise is incredibly distracting and I couldn't count how many times I thought everything is fine and then render finals and it's shit without details. Denoiser/or really clean render really shows how good or not the scene etc, is, while noise leaves lot for imagination.

[lacilaci]

Ok, so on twitter Attila Afra said that bidir without metropolis should be ok, I can't test now but is this possibe?

Check out @attila_afra’s Tweet: https://twitter.com/attila_afra/status/ ... 45859?s=09

[B.Y.O.B.]

I tested Bidir with Random sampler and the denoiser changed nothing.

[epilectrolytics]

I tested Bidir with Random sampler and the denoiser changed nothing.

Maybe also test Path metro?

[lacilaci]

I tested Bidir with Random sampler and the denoiser changed nothing.

Maybe also test Path metro?

path+metro works fine with nvidia. I think it would with intel too. Will test in the morning