#[solved]Artifacts with Smith Correlated Visibility Term and dark spots in PBR

12 messages · Page 1 of 1 (latest)

gilded bay
#

Hello. I am working on a Direct3D 12 renderer and got stuck with two PBR related issues. For now I do all testing for a single analytical point light and simple GLTF avocado model.

  1. I got black/dark spots on some angles. I went through PIX to check the values and when debugging the pixel I've think I narrowed down the problem to the values I got from LoH. For the dark spot it is really low causing the D term to be practically 0. (The dark spot screen I attached is lighted with two point lights but it is the same for 1 anyway).

  2. When I try to switch the implementation to smith correlated visibility term, I got really weird highlights that basically look like white artifacts. I have no idea why is that happening but I suppose it is also related to something being wrong with my halfway vector or view/camera direction. When I try to output the term directly, they both differ significantly (attached to images - the correlated one is the brighter one).

For both cases I have already checked values of other vector and NoL and NoV and they looked good. I attach both issues as screenshot and values from PIX of darkening.

I tried to provide as much info as possible. Appreciate any help, I am hard stuck as I checked all my matrices, vector and texture values. I don't understand why I get this artifacts.

smith_bug_3.png darkening.png correlated_bug.png uncorelated.png image.png 📎 pbr_loop.hlsl 📎 BRDF_implementations.h
snow quest
#

Do you mean NoH rather then LoH?
Also what’s the g_epsilon for? I haven’t come across that yet

gilded bay
#

Yeah, my bad for D term it is NoH and for G/V NoL and NoV, but the LoH is for Fresnel and the LoH value was low in PIX. g_epsilon is just a global const float of 0.00001.

#

Anyway I think I fixed black spots after a long 22 hour session of debugging. The dark spot is gone, I had a f*** typo in my camera position coordinates.

#

Now, when I switch to correlated I got this result

#

I am not sure about this small highlights happening at grazinig angles. Is it supposed to be happening for the correlated term?
This screenshots are from tonemapped rendering pass with ACES

#

I increased the roughness and it looks kinda weird still (the correlated one)

#

Well, at least the dark spots are gone.

gilded bay
#

[solved]Artifacts with Smith Correlated Visibility Term and dark spots in PBR