Incremental calc pass noiser than single frame...

[ShaunDon]

Hey everyone --

We just finished an animation for a client and we're looking to fix some glitches we simply didn't have time to address during production. The major one for me is the irradiance map -- I rendered single-frame tests before calculating incremental maps for each camera sequence, which rendered very clean, with only some soft noise and light leakage, but definitely tolerable for an animation.

The problem was that when I calculated the incremental map it generated a significantly greater amount of noise on the walls, especially on off-white surfaces. It takes about 18 hours for the irradiance map to calculate, so figuring out the solution will be potentially very time consuming for me -- can you all take a look at my results and my settings and offer any suggestions?

Thanks guys.

The rendered frame portions are left to right the single-frame result and the incremental result. The second has the direct sunlight turned off, but sunlight was on for both map calculations so it's an inconsequential change.





Thanks again,
Shaun

[percydaman]

why did you choose to put your Hsph subdivs at 10 and your Interp Samples at 50? Thats close to the exact opposite as the default. Try 80 for your Hsph and 30 for your Interp and see how that looks.

[ShaunDon]

Heh, pure trial and error. I'm still absorbing the manual. I was happy with the rendertimes and quality I was getting single-frame, I just don't understand the reason for the discrepency when I calc out the incremental pass.

[percydaman]

well its certainly an interesting problem. Ive found that its harder to get a noiseless animation when using single frame then using incremental add. But its a different type of noise then the one your exhibiting.

[Da_elf]

you sure it didnt have something to do with you hiding your light?

[ShaunDon]

I can render another test with the sunlight on, but with the irradiance map calc'd with the light on turning it off for the final render should only remove the direct light, which we composited into the final frame with a matte on top of the rest of our passes.

[ShaunDon]

Here's the version with the saved incremental map rendered with sunlight turned back on...



From what the manual says the only difference in settings recommended for animations is increasing the distance threshold (which you can see I haven't here) ... the way the manual describes the function -- "this parameter controls how sensitive the irradiance map is to distance between surfaces" -- didn't seem important in resolving this issue.

Also I should add that this is from frame 600 in the first sequence, after entering this atrium through an entry vestibule off-camera, so this room is revealed as the sequence starts... the only thing I can think of that would fix this is to somehow force VRay to reuse less information from the previous map than it is now. No idea if there is a way to control this.

Shaun

[vlado]

In fact, both images have the same amount of noise, but if you render single-frame, the -3/-2 setting for the Min/Max rate causes the samples to be relatively far away, and the noise is less obvious. When you use the Incremental Add mode in an animation, some areas (like corners or cracks) will get more densely sampled in the course of the animation than if you render single frame. Since the samples will end up close to each other, you will see the noise more clearly.

A solution would be to increase your hemispheric subdivisions for the irradiance map.

Best regards,
Vlado

[ShaunDon]

Thanks Vlado. I'll set that up to calculate tonight and hopefully that'll do the trick. I had planned to take a few days off now that this project's finally out, but it's so hard to leave it now...

Shaun

[percydaman]

Hey he only suggested what I already had...

[ShaunDon]

I realize that, percy... however I was more interested in a reason why that would make a difference in an incremental calc to justify the day it will take to test that idea.

However, I up'd the HSph subdivs from 10 to 30, which at least tripled the time it was taking to calc a single frame at 1024x512 (had been 28 minutes before with an 18-hour calc time for my first 1200-frame sequence at every 10th). A lot of pretty severe black splotches cropped up on the walls in the single-frame calculation, so I cancelled it after about 45 minutes.

Oh, my head hurts.

ShaunDon

[percydaman]

I was just making a joke...

[vlado]

A lot of pretty severe black splotches cropped up on the walls in the single-frame calculation, so I cancelled it after about 45 minutes.

That means that there is most probably another problem with your scene - double faces, for example.

Best regards,
Vlado