Support for decompressing to quantized (fixed-point) vertices

[kelleyma49]

As far as I can tell, draco doesn't support decompression into quantized formats (see https://github.com/KhronosGroup/glTF/tree/master/extensions/Vendor/WEB3D_quantized_attributes for more information.)

This would be great to have in order to minimize data transformation.

[bhouston]

So the idea is to not transform the quantized vertex positions and leave them quantized into the attribute buffers and then it is only in the glsl code that you transform them into floating point positions? This would allow one to have uint16 buffers rather than fp32 buffers for things like position attributes I guess?

[kelleyma49]

Correct. The transformation occurs in the vertex shader. Usually you transform the position attribute by a scale and offset passed in via shader variables.

You can also quantize texture coordinates and normals (see http://cesiumjs.org/2015/05/18/Vertex-Compression/ for an example.)

[ondys]

Draco supports compression of integer valued attributes, so it is possible to pre-quantize any attribute values before they are compressed by our library. At this moment, this would have to be done through our C++ API though. One drawback of this approach would be that the draco file would have no information about the quantization settings so these would have to be stored separately.

Skipping dequantization in our decoder would be certainly a more straightforward approach and we have been actually thinking about it already for exactly the use cases mentioned above. I've added it to our internal issue tracker and we will let you know when we have any update on this.

[ondys]

Support to skip dequantization in our decoder has been added in 1.0