Hello, I keep getting this noisy output from APV baking - how to mitigate that?:
I’ve tried to enable/disable Dilation/Virtual Offset but it seems have no impact on the final bake.
Baker settings:
Hello, I keep getting this noisy output from APV baking - how to mitigate that?:
I’ve tried to enable/disable Dilation/Virtual Offset but it seems have no impact on the final bake.
Baker settings:
Hoping there is a clear solution to the APV noise, maxing out the sampling and/or reducing probe spacing does not help.
In HDRP, SSGI helps hide the noise but seems like a costly band-aid fix.
I get the noise in interior scenes, exterior scenes, and like this one where I am trying to use sky occlusion with PBR Sky to make a dynamic day/night cycle while having interior locations in an open world.
Spatial noise like this usually caused by insufficient sample counts. Try cranking the the indirect and/or environment sample counts in the lighting window higher.
@Pema-Malling what’s the number you guys use internally to not have such nosie?
Also, isn’t what denoiser is for?
I got smooth results from light mapper with 256 sample count - this is too low for APVs?
S.
ok, so after some experiements, best result I got is:
No noise, barely noticable artifacts in this scene.
Sky occlusion is off, both validity options engaged, with default settings:
It also seems that just cranking up enviro settings is enough to get rid of the noise:
what’s the number you guys use internally to not have such nosie?
It really depends on the scene. The first scene you showed is mostly lit by indirect light from a small light source. Scenes lit mostly by indirect light tend to need more samples.
Also, isn’t what denoiser is for?
We don’t implement any kind of spatial denoising for probes. Denoising is exclusive to lightmaps.
I got smooth results from light mapper with 256 sample count - this is too low for APVs?
For this scene, seems like it. There are a number of factors contributing to that. For one, probes don’t get denoising. Also, lightmap texels only have to capture data from a hemisphere above the texel. Probes have to capture data from an entire sphere of direction. The space of possible light paths is larger, so you need more samples to adequately cover it.
It also seems that just cranking up enviro settings is enough to get rid of the noise
The environment sample count controls how much of the ray budget is used to sample the skybox. In the second scene you showed, most (all?) of the lighting is from the skybox. So makes sense.
Glad you figured something out
@Pema-Malling thanks for detailed explanation!