erich666
commented
2 years ago I made a replacement TestMap_Normal.png file, attached below.
Copied from https://github.com/KhronosGroup/glTF/issues/2223: Using NormalMap-Online, I negated the TestMap_blur.png image, negated the red channel (always needed), and set the z-range from -1 to +1 in the tool, then output.
Assuming Z ranges from -1 to 1: Vector 242 89 169 converts to 0.898, -0.302, 0.325, length 1.004
This compares well with the original range - see above.
This texture's Z values are properly in the -1 to +1 range, unlike the current versions in glTF-Sample-Models. I tested on the model TextureTransformMultiTest.gltf in directory glTF-Sample-Models/2.0/TextureTransformMultiTest/glTF with two viewers, Omniverse Create's and Windows' 3D Viewer. The results seemed reasonable. Replacing the same file in glTF-Sample-Models/sourceModels/TextureTransformMultiTest, the file TextureTransformMultiTest.blend looks reasonable in Blender.
DRx3D
lexaknyazev
I thought this was an issue with the glTF 2.0 spec itself, but the problem is simply that the normal texture TestMap_blur.png is poorly formed. There are a few of these PNGs used in this repo. They should be replaced with a proper texture, to avoid confusion (I certainly was confused to see it).
See https://github.com/KhronosGroup/glTF/issues/2223 for the full explanation and a possible replacement texture (one that is properly scaled in the blue, Z, channel).
Short version is that it maps blue [0.0 .. 1.0], not blue (0.5 .. 1.0], to z [0 .. 1]. For example, pixel 394,303 has a triplet of (242,89,82) - the blue value is too small, because they're using the wrong mapping. The spec is specific on this point:
The triplet is valid if mapping the whole range of blue values to z [0 .. 1], but that's not what the spec requires.
Here's a full analysis of this triplet. It's easy to convert with both mappings and see if the value (242,89,82) uses one of them.
Assuming blue (0.5 .. 1.0]: Vector 242 89 82 converts to 0.898, -0.302, -0.357 length was wrong at 1.012
Assume blue [0.0 .. 1.0] (which is the wrong mapping, not what glTF requires): Vector 242 89 82 converts to 0.898, -0.302, 0.322, length 1.001
Clearly this triplet is using the wrong mapping. When converted, it gives a vector of length 1.001 using this mapping, showing that this mapping is the intent of the texture.