What is the best way to encode coordinates in GLTF?

[Amoki]

BCF requires WorldCoordinates in meters for camera positions.

Meaning coordinates in Europe look like [3458214.879635, 5125369.158789, 12.125874]. Xeokit uses Float32 to handle coordinates and they are very imprecise with big numbers (up to ~40cm with coordinates around 5.000.000).

Which could be the best method to handle this: 1) Use Float64? I have no idea about the performances 2) Encode geometries with their world coordinates and quantization do some magic? 3) Encode geometries with local coordinates (small values easily handled by Float32) and set the offset in the root matrix. It does not behave well with Xeokit (the screenshot comes from this). We use a hack where we reset the root matrix to 0 before the load and apply the offet after the BCF export. 4) Encode geometries with local coordinates and define the offset in metadata. It's non standard. 5) Some other stuff

[Amoki]

I did some experimentation. Currently, we're doing stuff like:

const scene = gltf.scenes[0];
if (!scene) {
  return;
}
const nodeId = scene.nodes[0];
const node = gltf.nodes[nodeId];
const originOffset = [node.matrix[12], node.matrix[13], node.matrix[14]]
node.matrix[12] = 0;
node.matrix[13] = 0;
node.matrix[14] = 0;
gltfLoader.load({
  gltf: this.gltf,
  edges: true,
});

During getViewpoint() and setViewpoint(), I apply the offset. But it don't work with Xeokit format, it don't work if multiple models are loaded and they don't share the same origin.

I also tried

const originOffset = [node.matrix[12], node.matrix[13], node.matrix[14]]
gltfLoader.load({
  gltf: this.gltf,
  edges: true,
  position: [-node.matrix[12], -node.matrix[13], -node.matrix[14]]
});

The model is not bugged but the BCF viewpoint is still in Local Coordinates and not in World Coordinates. I don't know if I can set the position during GLTF-> Xeokit conversion. Multi models handling seems doable with this solution.

On the other hand, we don't need geometries to be in World Coordinates. Just the Camera needs it.

The needed features would be:

• [ ] getViewpoint() and setViewpoint() use World Coordinates
• [ ] Handle Xeokit Format
• [ ] Works with multiple models with different origins

[xeolabs]

This issue is related: https://github.com/xeokit/xeokit-sdk/issues/116

I need to prioritise client work at the moment, but I will be circling back to get in a solution to this issue soon.

[xeolabs]

This might help - I'm thinking of:

1. Adding a global realWorldScale property on the xeokit Scene, which is the number of real-world units (ie meters) per unit within the 3D world-space coordinate system.
2. Adding a global realWorldOffset 3D vector property on the xeokit Scene, which is the distance of the xeokit world-space origin from the real-world origin, in real-world units (ie meters).
3. Adding realToWorld(coord3d) and worldToReal(coord3d) utility methods to Scene to help map back and forth between real-world units and xeokit world-space coordinates.

Then the BCFViewpointsPlugin setViewpoint and getViewpoint methods could use those methods somehow to map to and from real-world coordinates.

Does that sound like a useful approach?

[Amoki]

The realWorldScale doesn't seem that useful in the BCF use case. realWorldOffset and related methods are good! Unfortunately, this is a custom method and nothing append in the GLTF but if there is no other way, it is the right way!

[xeolabs]

Just a heads-up that I've almost finished ability to position/scale/rotate XKT models as we load them. Just one small BIMServer-related bug to fix. Will be in master next week.