render differences between tiles

Started by Kevin F, July 02, 2007, 03:16:29 AM

Previous topic - Next topic

Kevin F

The render differences that are apparent when you split a file into two halves and render independently, can also be seen in the same scene between rendered tiles.  I've marked where they appear on the attached image which has just finished rendering after 59hrs!
The settings on this were deliberately high to remove the grainy clouds visible on other tests at lower settings.
Settings were: detail .8, AA 5, GI 2,2, atmosphere quality 64 samples, cumulus 500 samples, (Oshyan, I tried lowering this to 300 samples but the graininess was still there). The two barriers on the right are not too bad but the other two are very noticeable and distracting. Are these artifacts a result of the high settings or are they just specific to this scene? I'd have thought that within the same render session this wouldn't happen. Also, on a different issue, what makes the bright area of clouds to the upper left look like melting ice cream?
Any Ideas?
Regards
Kevin.

bigben

The differences are a known issue with GI. GI is only calculated on the rendered portion of the image and isn't "world" aware. It is particularly noticeable in areas of shadow.

Clouds also look a bit different at higher detail settings from what I've tried. The "melting ice cream" is the result of the combination of your cloud settings. If you tweak the cloud settings with detail turned down it's a good idea to at least preview a small cropped section with higher detail, particularly near the edges.

Kevin F

Quote from: bigben on July 02, 2007, 11:20:15 PM
The differences are a known issue with GI. GI is only calculated on the rendered portion of the image and isn't "world" aware. It is particularly noticeable in areas of shadow.

Clouds also look a bit different at higher detail settings from what I've tried. The "melting ice cream" is the result of the combination of your cloud settings. If you tweak the cloud settings with detail turned down it's a good idea to at least preview a small cropped section with higher detail, particularly near the edges.

Thanks for your input Ben. I know TGTP has GI issues between split renders (two sessions) of the the same TDG file, but within a single render between rendering tiles? Where is this mentioned? That means that every render with GI switched on can give different versions of the same "world" file for each rendered tile! Is this true Planetside? Wherever you are.

Oshyan

As Ben said GI is only calculated for the rendered area (as an obvious way to save time rendering). It of course takes into account "off screen" direct light, but indirect light bounces can't be accounted for without calculating the entire scene. This can result in anything from subtle to quite dramatic differences between adjacent renders of the same scene as well as crop render areas of the same render. The principle is the same - it's only taking into account what is in the rendered area.

This is certainly something that we will be working to address but due to the GI system operating on such a wide area it may be difficult to fully eliminate. Indoor-focused GI solutions have different challenges and have received more attention by mainstream developers so there are more ready solutions.

Btw in regards to high cloud vs. atmosphere samples, in situations like yours of heavy shadows and rays the atmosphere samples will tend to have a much greater impact on noise. In fact you can still see a little bit of noise evident even at 500 samples, and this is almost certainly due to the atmosphere sampling level. You might find equivalent or better results in similar or less render time with 300 cloud samples and 128 atmosphere samples.

- Oshyan

Kevin F

Thanks Oshyan,
We live and learn.
Useful info re the cloud/atmos settings - thanks.
Regards
Kevin.

Matt

While most of what Oshyan said is correct, I should make a small correction here.

Quote from: Oshyan on July 03, 2007, 04:10:24 PM
...indirect light bounces can't be accounted for without calculating the entire scene ... it's only taking into account what is in the rendered area.

Indirect light bounces from the rest of the scene are accounted for. When you render a cropped region, indirect light bounces arriving at the cropped region are simulated, no matter where in the scene the light comes from. Nevertheless, there are problems with cropped regions, and the cause is due to inadequate GI samples at the edges of the crop region and the way the GI samples are interpolated. Currently there are not enough samples from areas slightly outside the cropped region to ensure consistency at the boundary. It's particularly difficult because the interpolation covers quite a significant area of the image in order to prevent "blotchiness". I do think I can solve this problem in future.

Matt
Just because milk is white doesn't mean that clouds are made of milk.

Volker Harun

So it could be possible to prevent this issue by rendering overlapping crops and blend them?

Kevin F

Quote from: Matt on July 05, 2007, 09:20:34 AM
While most of what Oshyan said is correct, I should make a small correction here.

Quote from: Oshyan on July 03, 2007, 04:10:24 PM
...indirect light bounces can't be accounted for without calculating the entire scene ... it's only taking into account what is in the rendered area.

Indirect light bounces from the rest of the scene are accounted for. When you render a cropped region, indirect light bounces arriving at the cropped region are simulated, no matter where in the scene the light comes from. Nevertheless, there are problems with cropped regions, and the cause is due to inadequate GI samples at the edges of the crop region and the way the GI samples are interpolated. Currently there are not enough samples from areas slightly outside the cropped region to ensure consistency at the boundary. It's particularly difficult because the interpolation covers quite a significant area of the image in order to prevent "blotchiness". I do think I can solve this problem in future.

Matt


Thanks for that Matt.
So, let me get this straight: adjacent tiles in a single render session of an image behave in the same way as adjacent cropped areas of a scene, rendered in different sessions?
Kevin.

Oshyan

Thanks for the correction matt. ;D

The problem should be somewhat improved by doing overlapping crops/renders, yes. The problem occurs in crop renders and adjacent render tiles (for a panorama for example) regardless of being in the same or different sessions. The GI data is not cached at this point.

- Oshyan