Ben Supnik Wrote:Yep - this is exactly what I saw when I had triangulation on. But I think in theory a weighting by arc-angle could fix this. Two of the three triangles at a given vertex will share 90 degrees of "span" and the other will have 90 to itself - by this weighting, normals will go back to vertical. I don't know what the effect would be on other shapes.

When implementing LDView's smoothing, I actually considered using the un-normalized cross product as the input to the average calculation. Since the magnitude of the cross product is proportional to the area of the triangle, I think this would result in correct surface normals in the case of a cylinder. However, in order for it to work, quads would in fact have to be split into triangles, and my smoothing code acts on the quads.