Don't Affect Colors & adaptive subdivision

[Shimakaze]

Is it just me or doesn't this function work as it should?
Whenever I use it with adaptive subdivision the rendertime doubles.
Adaptive DMC seems to work fine though.

Here's an example:

All pictures rendered with default settings: Irrmap High, Light Cache defaults, Adaptive Subdivision, Linear Multiply.

Gamma 1.0 (rendertime 34,9 seconds)


Sample Rates:


Gamma 2.2 (rendertime 40,3 seconds)


Sample Rates:


Gamma 2.2, Don't affect colors (1 minutes 7,3 seconds)


Sample Rates:


From my understanding the third picture should have been rendered with the same sample rates as the second (without affecting the colors) but this is obviously not what happens. It also takes much longer.

[Shimakaze]

...Anyone?

[ene.xis]

i haven t had time to test this myself!
i was hoping vlado to clear this up...so here goes another <bump>

[pg1]

I havent had time to test it either was hoping it would work perfect and I would never have to worry about noise again

maybe this helps

http://www.chaosgroup.com/forums/vbu...ad.php?t=41048

Will try and look into it today

[Aldaryn]

Is it just me or doesn't this function work as it should?(...)

It's actually is how it should work. With the "don't affect colors" on, gamma correction will be only taken into account during sampling. So, this switch leaves colors un gamma corrected, but helps the sampler to adapt to a gamma level different from 1.0. This is very useful, if you write an image with linear output, but eg. sRGB display (exr's for example) so you will not burn a gamma correction into the output, but the samplerer will be able to guess final tones better, especially in darker areas and midtones.
Just try to apply a gamma correction of 2.2 in post to your output, and you will see the difference. And, of course, if you don't see a difference in the linear images proves how well the adaptation of the samplerer works.

Best regards,
A.

[pg1]

sorry the thread I linked to wont be any help

Just looking at the images again, Im not sure what the issue is the images are correct with what should happen.

Adaptation only applys a gamma correction what vray will antialias (the samples element) not the colours i.e all the other elements (exactly like the name implies)

As I understand it the reason for using it is that as the gamma curve isnt linear vray will antialias more in the darks removing noise from the shadow areas as you will be brightening the image in post.

As there is now a higher difference between the colour of the pixels (in the now gamma corrected samples) Vray will anti alias more and take longer to do it.

That is exactly what your images show.
OPPS did see you there aldaryn you beat me to it

Just to add I have no idea why you would get a different result with adaptive DMC

[Shimakaze]

The problem is when rendering an image with the don't affect colors option enabled it actually uses a whole lot more samples then it's supposed to.
The actual samples should be the same in the second and third picture.
The second picture is rendered with baked in Gamma 2.2 and the third is rendered with the don't affect colors option, leaving it in Linear space (although I afterwards gamma corrected them to be identical).
By the way in case I didn't make it clear before, the normal pictures and sample rates belong together. First the normal picture, then the sample rate for that picture and so on.
I really doubt rendering with the don't affect colors option is supposed to double rendertime compared to an image rendered with the same color mapping setting (which means the samples should be the same as well).
And as for the DMC sampler, when doing the exact same renders with it the second and third pictures actually do have the same sample rates and speed so it seems obvious that's how it's supposed to work.
Let me stress this... there's no problem with the picture quality. it's the rendertime and amount of samples being used that's not right.

[Aldaryn]

No, not at all. If checked, sampling will adapt to a gamma corrected image, but the output won't have gamma baked in. It is this simple. Really. And since you've got brighter midtones with a larger gamma, the samplerer will take more samples to maintain the specified noise level in a gamma corrected image's aspect. It's really not a but. This is what this switch does.

Best regards,
A.

[ene.xis]

Aldaryn if i am not mistaken* Shimakaze agrees and understands that what he is pointing out is that the rendertimes of a linear rendered image with the don t affect colors option turned on and and another with the gamma baked in should not be 2x nor the sampling that different imHo.

*of course i have a tendency to be wrong and the daily 5 hours sleep of the last weeks might routine might not be helping me neither

[vlado]

In fact, the sampling can be very different, if the original image is relatively dark. As the colors go towards the zero, a small change in the original image becomes a much larger change in the gamma-corrected image. However there shouldn't be much of a difference between a rendering with the "Don't affect colors" and without it. There was a bug with versions of V-Ray with regards to this, but it was supposed to be fixed for SP2.

Best regards,
Vlado

[Shimakaze]

ene.xis: Indeed

Vlado: Yea if you look at the second and third picture both are in fact rendered with the same color mapping (the first is not) except for the don't affect colors option and they should be similar (as they are when using Adaptive DMC) But they are wildly different and the rendertime goes through the roof.
Dunno if it's the same bug as before or a brand new one though?

[ene.xis]

i guess some small part of me brain might be working after all!

[Shimakaze]

Ene.xis: Or wait... hmmm I'm confused now as to what you ment...
The rendertime of the second and third picture should be different from the first, but not that different from each other.

[ene.xis]

i mean rendering a linear image with dont affect colors on and rendering an image baking the gamma should not be twice as much...

[Shimakaze]

Yea. Since both pictures are treated as having gamma 2.2 even though the picture with the don't affect colors option turned on is output as linear, they should both have similar sample rates. The time difference in that test wasn't that big (40s vs 1m7s), but in more complex pictures I've had the same render with the exact same setting take 3 times longer with the don't affect colors option on.