Bug(s) in Rust port

[markusmoenig]

Hi,

I have ported your path tracer to Rust (https://github.com/markusmoenig/rust-pathtracer) so that I can use it in the 3D modeling programming language I am working on (forgedthoughts.com).

However there is (or are) bug(s) in the port. For example a value of 0.0 for roughness just creates black, non-reflective materials. I went over the whole port 1-2 times and couldn't find anything.

Do you have any pointers in the code where I should focus on for the above behavior ? Thanks!

[knightcrawler25]

Hey :)

A roughness value of 0.0 requires special handling of the sampling routines due to the delta distribution that's involved.

I avoid this issue by clamping the min roughness value to 0.001 when bringing the material data into the shader here: https://github.com/knightcrawler25/GLSL-PathTracer/blob/e1069a5379ab4aaf9fe4fe8ad6d5ee402da0c652/src/shaders/common/pathtrace.glsl#L46 and here: https://github.com/knightcrawler25/GLSL-PathTracer/blob/e1069a5379ab4aaf9fe4fe8ad6d5ee402da0c652/src/shaders/common/pathtrace.glsl#L110-L111

Not doing that will cause the material to appear black as you describe:

This is with the clamped roughness:

[markusmoenig]

Sorry. I should have clarified. I clamp the roughness but it's near black. Metals are also very dark. Have to have light sources with 100 emissions to see reflections. Somewhere something is wrong.

[knightcrawler25]

At first glance, one thing that seems to be missing is when the ray hits a light source by chance, power_heuristic should be used to combine the bsdf sample and the light sample for MIS. Its being done in direct_light() while sampling light sources but is missing in render()

This happens here in my code: https://github.com/knightcrawler25/GLSL-PathTracer/blob/e1069a5379ab4aaf9fe4fe8ad6d5ee402da0c652/src/shaders/common/pathtrace.glsl#L348-L363

I'm new to Rust so trying to figure my way through the code :D Will see if I can narrow it down further

[markusmoenig]

Wow thanks for your effort!! Will look into it tomorrow. Already quite late here in TH…

[markusmoenig]

I have implemented the missing light sampling and the call to power_heuristic.

Things look good now (I think). The example scene now looks like this. Thanks again for your help! Still need to implement some stuff (env_maps and I am not sure about mediums and how to render them for SDFs) but will do so as I need them for ForgedThoughts.

[knightcrawler25]

Glad to see that the issue is fixed now :)

Also, the Disney BSDF in your port is from an older version of the code which had some issues with interiors/backfaces as mentioned here: https://github.com/knightcrawler25/GLSL-PathTracer/issues/78

Depending on how you're mixing the transmission parameter, you might end up seeing similar problems in very specific scenarios. However, it's been fixed with the most recent commit.

[markusmoenig]

Ok, will try to keep the code up to date and update it when new releases come out. Thanks for the pointer.

[markusmoenig]

Closing this for now.