Energy Inconsistency with local sampling

[gustavoeb]

Hi,
In daily use of the Mental Ray renderer I noticed a energy inconsistency while playing with local vs global sampling... I was quite bogused by this and decided to test it out a little further.
You can see by the attached imaged that:
(1) Energy is higher with 1 eye ray then with 64 eye rays per pixel
(2) When there is 1 ray per pixel and varying number of local glossy reflection samples energy stays consistent
(3) When there is 64 rays per pixel energy increases as local reflection samples increase :

energyInconsistency.jpg

[Remydrh]

Out of curiosity, separate this into direct (specular) and indirect reflection. We found previously the result change was in the direct reflection component (specular).

Is that true here?

[gustavoeb]

This is the reflection of a visible Portal Light.

[Remydrh]

Yes, but the shader you are using, I assume mia, also has a specular component. Render to passes and compare the change in specular versus reflection. In our experience the reflection is consistent, but the specular amount changes.

I wanted to know if this was true in your tests as a confirmation.

[gustavoeb]

I would expect direct reflections of visible portal lights to show up in Specular buffer, but they dont. They are actually in the reflection buffer.
To get rid of any doubt this is the same test but with both a portal light and emissive geometry, results are the same.

geoVsLight.jpg

[Remydrh]

I tried this with Unified Sampling, min 1 max 800 quality 6

mia reflection samples 1, 32, and 64

I did not see this difference, they were all the same. Have you tried this with Unified Sampling? Perhaps I'm missing something you're doing. This is a geo plane and reacts as expected to reflection samples. Previously observed was regular lighting for area lights (not portal) seems to alter the spec intensity when changing samples for the area light.

Attached are 1 and 64 samples.

1sample_reflection.jpeg64sample_reflection.jpeg

[fatjulio]

I've also noticed this. Are you using a lens shader to correct the gamma? It seems the lens shader is what's causing the problem. It does this with soft shadows as well.

Without a lens shader, the results are the same whether you put the samples locally, or globally. With the gamma correction node, the local samples version is correct, but the global samples version is not. It doesn't matter if it's unified sampling or old style anti aliasing.

[Remydrh]

Hm, might need to try that. Certainly worth straightening out.

I do not color correct through a lens shader. In fact, you should not. This should be done in post. It's provided as a preview convenience. Use the built in tonemapping of something like Nuke or the Maya Render Window to view the image when rendered to 16-half (float) EXR or similar.

[gustavoeb]

Yes it happends to any type of sampling in MentalRay, as I described in the beggining I'm using fixed (unbiased) sampling as I thought this my be related to the adaptation of samples (now I know it does not.

To reproduce this easier you might have to use High emissive values and very diffuse glossy reflections.

I am NOT using a lens shader for color correcting, though I dont know how the 3d package I'm using (Softimage) is doing this internally. I'm using the correct worflow inside the software, and I suppose it would be color correcting the final image (which would not be a problem). This image was saved in an 8bit format with gamma baked in.

Will render a linear EXR tomorrow to see if I get correct results. If so then the DCC is to blame, lets see

[gustavoeb]

Mistery solved!
When rendered out to EXRs this work perfectly. With ALL sampling methods in MR, very consistently...

This is related to some problem in the Render Region inside Softimage and the way it color corrects the image.