Incorrect decoding of draco model with high quantization bit number

[OmarShehata]

This was reported by Chris Wang on the forum.

The glTF file attached in this post was encoded with the following Draco parameters:

sOpt.pos_quantization_bits = 20;
sOpt.tex_coords_quantization_bits = 15;

And renders incorrectly in Cesium. Here's what it looks like in Bablyon:

And in CesiumJS:

@lilleyse believes this might be because the quantization is hardcoded in a few places in our Draco decoding.

[lilleyse]

@lilleyse believes this might be because the quantization is hardcoded in a few places in our Draco decoding.

Specifically I wouldn't be surprised if UNSIGNED_SHORT is hardcoded somewhere.

[wangzhongliang]

@lilleyse Hi Sean, will this be fixed soon? if not, can you tell me where is possibly hardcoded, so that I can fix myself.

[wangzhongliang]

When I set dequantizeInShader to false, the model is not distorted, and displays well.
Hope it will help.

[lilleyse]

This also came up in CesiumGS/gltf-pipeline#511

[lilleyse]

Opened an issue in the Draco repo about whether it's possible to query the quantization bits without decoding to quantized form: google/draco#573

@sanjeetsuhag and I talked offline - if there's no way to do it in the API, we'll need to dequantize in our own code.

[Ben-Mack]

@lilleyse Has this bug been fixed on the latest Cesium?

[lilleyse]

@Ben-Mack, no, still an open issue

[lilleyse]

Came up again in #10516

[lilleyse]

Came up again in #10977

[lilleyse]

Fixed in #10979