Differential D12403
Fix T91064: Cycles low poly meshes having black edges when shade smoothed Authored by Mikhail Matrosov (ktdfly) on Sep 5 2021, 1:17 AM.
Details
Fixes:T91064: Cycles Lower Poly Meshes Having Obvious Black Edges When Shade Smoothed Caused by rBcd118c5581f4: Fix T90423: black pixels after shadow terminator geometry offset
Diff Detail
Event TimelineComment Actions Looks good, I'd just like an explanation for the *pdf = 0.2f; so that I understand it.
Comment Actions The pdf value of 0.2 makes little sense to me. It's also not clear to me that doing this in the principled diffuse BSDF is enough, we also have the regular diffuse BSDF and other BSDFs. My understanding is that what happens here is that we end up with a bumped/smooth normal such that the view direction lies outside of the hemisphere around that normal. I think it may be better to use ensure_valid_reflection and adjust the normal to avoid that situation, and do it for all BSDFs instead of having each BSDF evaluation solve it on their own. Comment Actions Adds ensure_valid_reflection(). Minor optimizations to the Diffuse BSDF I'll post as a separate patch. Comment Actions This revision looks good but I think you should keep your numerical change (I highlighted it in the comment) as it is way better than the current implementation IMHO.
Comment Actions The results look great and it appears to work. You just need to set the pdf value at the comment and this is ready to go.
Comment Actions
In my experiments it was needed only for the case where dot(N, V) <= 0, but since N is ensured to be valid, it is not anymore. Comment Actions Can we do ensure_valid_reflection only for triangle primitives? I'm seeing some changes in hair, but I don't think that needs to change since the shadow terminator offset is only for triangles. Otherwise the results look good to me. There is a bit of a performance hit but I'm not seeing a way around it. Comment Actions
| ||||||||||||||||||||||