A bump-mapped cornell box in PBRT back

(L) [2012/09/11] [ost by waldheinz] [A bump-mapped cornell box in PBRT] Wayback!

I'm currently playing around with pbrt and found something I don't understand. I did my experiments with the path tracer. The first image shows a pretty much standard cornell box, where the green wall has some FBM bumpmap applied. It looks pretty much as I'd expect it: The play of light and shadows on the bumpmapped wall gives an indication where the light source is.

[IMG #1 Image]

Then I modified the path tracer to disable the next event estimation, by removing the (depth == 0 // specular) test when adding the light emitted at the intersection and removing the call to UniformSample... . So the ray now will bounce around until it finally hits the light by chance. I expected the resulting image to be the same as the original one. Se for yourself:

[IMG #2 Image]

The appearance of the wall is completely changed. But why? If I understand it correctly, we don't need the adjoint for either of the images. But I found that it seems lots of rays coming from the camera are "reflected" to the backside of the green wall, instead of back into the scene. But honestly I'm having a hard time deciding if that's ok, given that bump maps aren't physically based anyway.

PS: Please excuse the crappy tonemapping.

---

[IMG #1]:
[IMG #2]: