New Rough Radiance Approximation #16063
Conversation
|
Please make sure to label your PR with "bug", "new feature" or "breaking change" label(s). |
packages/dev/core/src/Materials/Node/Blocks/PBR/pbrMetallicRoughnessBlock.ts
Outdated
Show resolved
Hide resolved
| @@ -1255,6 +1255,13 @@ export class PBRMaterialPropertyGridComponent extends React.Component<IPBRMateri | |||
| propertyName="useHorizonOcclusion" | |||
| onPropertyChangedObservable={this.props.onPropertyChangedObservable} | |||
| /> | |||
| <CheckBoxLineComponent | |||
There was a problem hiding this comment.
Triggering this doesn't change the shader right now. I think I'm missing something to make this property change force the shader to recompile.
There was a problem hiding this comment.
Just replace:
if (this._useAlternateRoughRadiance) {
defines.ALTERNATE_ROUGH_RADIANCE = this._useAlternateRoughRadiance;
}with
defines.ALTERNATE_ROUGH_RADIANCE = this._useAlternateRoughRadiance;in pbrBaseMaterial.ts.
|
Snapshot stored with reference name: Test environment: To test a playground add it to the URL, for example: https://snapshots-cvgtc2eugrd3cgfd.z01.azurefd.net/refs/pull/16063/merge/index.html#WGZLGJ#4600 Links to test babylon tools with this snapshot: https://playground.babylonjs.com/?snapshot=refs/pull/16063/merge To test the snapshot in the playground with a playground ID add it after the snapshot query string: https://playground.babylonjs.com/?snapshot=refs/pull/16063/merge#BCU1XR#0 |
|
Visualization tests for WebGPU (Experimental) |
|
WebGL2 visualization test reporter: |
MiiBond
force-pushed
the
mbond/rough-radiance-improvement
branch
from
9c5117a to
c0bb1e5
Compare
|
Hey @MiiBond! Is there a physical argument for lerping radiance to irradiance with higher roughness values, or is it simply because the results look better? Also, have you been able to test with larger scenes / other HDR files to compare with the existing code path? |
So, the short answer is that it's not physically correct to essentially substitute diffuse reflection for specular at roughness levels near 1.0 (they aren't the same thing). However, the reason that it looks more plausible in cases like what I've shown above is that the importance sampling for radiance at roughness near 1.0 really starts to do a very poor job when an IBL has small, strong light sources. Basically, the importance sampling is centered around the reflection direction but doesn't take into account the intensity of the IBL in a given direction. So, on the rough end, you end up with a blurred reflection without the intensity of strong light sources taken into account. This is why we lose the terminators for the strong lights. We might be able to improve the filtering of radiance at the rough end using the CDF maps but I'm not sure how exactly to do that and I haven't had time to explore it.
Yes, I've been using this approximation in Substance Stager for a while now. I can post some more comparisons in a minute. |
|
These include both the improved irradiance sampling (using CDF) and the new roughness approximation: |
|
|
|
Thanks a lot for the explanation and the screenshots! I will review the PR asap. |
|
Thanks a ton for the explanation !!! |
packages/dev/core/src/Materials/Node/Blocks/PBR/reflectionBlock.ts
Outdated
Show resolved
Hide resolved
packages/dev/core/src/Materials/Node/Blocks/PBR/reflectionBlock.ts
Outdated
Show resolved
Hide resolved
packages/dev/core/src/ShadersWGSL/ShadersInclude/pbrBlockReflection.fx
Outdated
Show resolved
Hide resolved
packages/dev/core/src/ShadersWGSL/ShadersInclude/pbrBlockReflection.fx
Outdated
Show resolved
Hide resolved
There was a problem hiding this comment.
Can we move the new setting in PBRBRDFConfiguration.ts ? as it is really advanced ? Also, should we turn it on by default ? I quite like the visual improvment. We could revert the value to false if we consider it lowering the quality in some given cases ? And finally, is it also valid when relying on an irradiance texture instead of SH as it could have more dynamicity ?
MiiBond
force-pushed
the
mbond/rough-radiance-improvement
branch
from
793bdad to
afeb399
Compare
Yes, I really like this idea. It's cleaned up the added code a fair bit and I think it makes sense to enable this by default. |
|
Visualization tests for WebGPU (Experimental) |
|
WebGL2 visualization test reporter: |
The way prefiltered radiance is applied in rasterizers tends not to match the raytraced ground truth very well towards the rough end of the spectrum. I've found that mixing with the irradiance gives a much better result as you can see in the images below.
Here's a playground: http://localhost:1338/#RNO672#1
Raytraced:
Babylon:
