Implement SubstrateMaterial

[wahn]

The test scene assets/scenes/ganesha/ganesha.pbrt uses a substrate material:

$ rg substrate assets/scenes/
assets/scenes/ganesha/ganesha.pbrt
38:    Material "substrate" "texture Kd" "tmap" "color Ks" [0.04 0.04 0.04]

On the C++ side we have this class:

// SubstrateMaterial Declarations
class SubstrateMaterial : public Material {
  public:
    // SubstrateMaterial Public Methods
    SubstrateMaterial(const std::shared_ptr<Texture<Spectrum>> &Kd,
                      const std::shared_ptr<Texture<Spectrum>> &Ks,
                      const std::shared_ptr<Texture<Float>> &nu,
                      const std::shared_ptr<Texture<Float>> &nv,
                      const std::shared_ptr<Texture<Float>> &bumpMap,
                      bool remapRoughness)
        : Kd(Kd),
          Ks(Ks),
          nu(nu),
          nv(nv),
          bumpMap(bumpMap),
          remapRoughness(remapRoughness) {}
    void ComputeScatteringFunctions(SurfaceInteraction *si, MemoryArena &arena,
                                    TransportMode mode,
                                    bool allowMultipleLobes) const;

  private:
    // SubstrateMaterial Private Data
    std::shared_ptr<Texture<Spectrum>> Kd, Ks;
    std::shared_ptr<Texture<Float>> nu, nv;
    std::shared_ptr<Texture<Float>> bumpMap;
    bool remapRoughness;
};

We have to implement the Rust counterpart ...

[wahn]

The C++ version renders the scene like this:

[wahn]

No progress on the material itself, but the graphics state has a default (matte) material now (used for the ground triangles), so we can (at least) render with Rust:

[wahn]

To start working on the material itself we first had to take the reverse_orientation boolean into account. See commit dd1f62a. Otherwise we do not see any contribution from the light in front of the detailed scan of a small statue:

> imf_diff -d -f ganesha.exr pbrt_rust.exr diff.jpg
differing pixels:	  2.981% (15455 of 518400)
average difference:	  1.251%
maximum difference:	  4.918%
Summary: Some pixels differ slightly.
== "ganesha.exr" and "pbrt_rust.exr" are similar

[wahn]

The last commit 1aeb3ec somewhat implements most of the material properties:

Beside the different lighting (which probably comes from replacing the infinte light sky texture through an HDR version of the OpenExr image) there seems something wrong with either the uv-coordinates of the mesh, or the texture being used on the statue is missing entirely ...

[wahn]

Commit 590fc48 fixes the different lighting by multiplying the read texture pixel values by a given scalar value:

So, now let's focus on the texture problems for the statue itself ...

[wahn]

Commit 1aead72 fixes the issue and the resulting image is very similar to the C++ version (which BTW uses OpenEXR for lighting vs. the Rust version being used for the comparison uses HDR):

> imf_diff -d -f pbrt_cpp.png ~/git/github/rs_pbrt/pbrt.png diff.jpg
differing pixels:	  6.341% (32870 of 518400)
average difference:	  1.222%
maximum difference:	  3.287%
Summary: Many pixels differ slightly.
== "pbrt_cpp.png" and "/usr/people/jan/git/github/rs_pbrt/pbrt.png" are different

Closing the issue ...