Microfacet BxDF with small roughness does not match specular BxDF

[alekseyvo]

Hi,

I expect the glass material with very small roughness to almost exactly match the glass material with zero roughness (while the first one uses microfacet BSDFs and the second one uses specular BSDFs). However, the difference is quite visible.

In this example scene left sphere has roughness 0.0 and right sphere has roughness 0.000001. Note the left sphere has clearly visible highlight on top while the right one misses this highlight.

Integrator "path" "integer maxdepth" [ 65 ] 
LookAt 0 4 -10  0 0 0  0 1 0
Sampler "stratified" "integer xsamples" [ 100 ] "integer ysamples" [ 100 ] "bool jitter" [ "true" ]
Film "image" "integer xresolution" [ 640 ] "integer yresolution" [ 360 ] "string filename" [ "scene1.png" ] 
Camera "perspective" "float fov" [ 45 ] 
WorldBegin
    AttributeBegin
        AreaLightSource "diffuse" "rgb L" [ 4.0 4.0 4.0 ] 
        Material "matte" "rgb Kd" [ 0 0 0 ]
        Shape "trianglemesh" "integer indices" [ 0 1 2 0 2 3 ] "point P" [ -4 8 -4   4 8 -4   4 8 4    -4 8 4 ]
    AttributeEnd

    AttributeBegin
        Translate 3 1 0
        Material "glass" "rgb Kr" [ 1.0 1.0 1.0 ] "rgb Kt" [ 1.0 1.0 1.0 ] "float eta" [1.5] "float uroughness" [0.000001] "float vroughness" [0.000001] "bool remaproughness" [ "false" ]
        Shape "sphere" "float radius" [2.0]
    AttributeEnd

    AttributeBegin
        Translate -3 1 0
        Material "glass" "rgb Kr" [ 1.0 1.0 1.0 ] "rgb Kt" [ 1.0 1.0 1.0 ] "float eta" [1.5] "float uroughness" [0.0] "float vroughness" [0.0] "bool remaproughness" [ "false" ]
        Shape "sphere" "float radius" [2.0]
    AttributeEnd

    AttributeBegin
        Texture "checks" "spectrum" "checkerboard" "float uscale" [100] "float vscale" [100] "rgb tex1" [1 1 1] "rgb tex2" [0.98 0.98 0.98]
        Material "matte" "texture Kd" "checks"
        Shape "trianglemesh"  "integer indices" [0 1 2 0 3 2 ] "point P" [50 -1 -50    -50 -1 -50    -50 -1 50    50 -1 50 ]
    AttributeEnd
WorldEnd

[TzuChieh]

IIRC, pbrt uses a function to re-map user specified roughness values. The roughness value used for BSDF evaluation will not match the one you specified. I'm not sure, however, that it can cause such a big difference in rendered images.

[mmp]

The issue seems to be that total internal reflection isn't handled properly (cf. #254). Fix forthcoming.

[mmp]

Fixed now. This scene is a good test case--I've added it to my little pbrt test suite.