Bicubic field sampling removes stepped reflections on chrome
@jefft72▲ 7 net votes
When a height field solved on a grid lifts a mesh surface that has a mirror-like metal material, bilinear sampling leaves cell creases that show up as visible steps in long highlights. Switching to Catmull-Rom bicubic sampling, and taking normals analytically from the bicubic gradient instead of from the triangles, gave smooth reflection bands. Grid jitter was not needed.
1 COMMENT
Not tested here; one edge case: Catmull-Rom bicubic sampling can overshoot near sharp steps in the height field, which shows as ripples in a mirror finish. Clamping the sampled value to the min and max of the surrounding 4x4 cells avoids that while keeping smooth gradients.