3D Tiles fixes and perf improvements

[jailln]

Some 3D tiles fixes:

• Fix a bug that impacted display of 3D tiles with sphere bounding volumes, closes Unstable 3d tiles rendering in itowns #1859
• Fix display of 3D Tiles bounding volumes of type sphere, closes 3D Tiles debug tools: bounding spheres are not displayed #1888 and Cesium has more detailed tiles texture render on zoom #1853

Perf improvement:

• Improve perfs for 3D tiles with bounding region by converting bounding regions to bounding spheres. Previously, bounding regions where converted to OBB and the AABB englobing the OBB was used for culling and SSE computation. Now, they are converted to bounding spheres and the sphere is used for culling and SSE computation. Depending on the cases it can be a bigger approximation than before but it is faster and I didn't see any visual discrepencies. I tested the perf differences on a 3D Tiles photogrammetry tileset (~500Go, 13732 tiles with only bounding regions) that I cannot share unfortunately. On the master, we spend most of our time initializing the bounding region:

After the PR, we spend most of the time in parsing and preparing for rendering:

I also tried to implement a conversion to threejs' OBB and to implement a culling algorithm on OBB (SSE computation for OBB is easy) but didn't succeed completely (can be seen on the following branch). Another way would have been to improve OBB.setFromExtent which is the culprit but I haven't found another way of implementing it yet. Propositions are welcome :)

BREAKING CHANGE: Remove region, box and sphere properties of C3DTBoundingVolume. They have been replaced by volume property which contains a THREE.Box3 (for box) or a THREE.Sphere (for sphere or region). Initial bounding volume type can be retrieved with the initialVolumeType property.

[valentinMachado]

This review does not treat the solution implemented (replacing C3DTBoundingVolume by sphere) and if this is actually working.. but is more about the code in itself.

What could I do to check if the solution is working ?

[valentinMachado]

I think it's a poor pattern to store in a single variable different type of object (volume) then to use another variable (here initialVolumeType) to know how to manipulate volume.. (btw you could use instanceof instead of initialVolumeType)

you could create a wrapper abstract class called Volume implementing a method isVisible(camera, tileMatrixWorld) then specifying SphereVolume and BoxVolume (class extending Volume) the specific behavior.

you could also put two attributes in C3DTBoundingVolume instead of volume one called box and the other one sphere which is the solution I rather, since a C3DTBoundingVolume could have the both and it avoids to create an abstract class

[jailln]

The last pattern was the previously used pattern. However, if we choose this one I would keep the initialVolumeType to be able to know if the sphere comes from a bounding sphere or from a bounding region, in case it may be useful for itowns' users in some way. The second pattern is also interesting, I actually thought about it when implementing this. I don't have time to implement it right now but I'll come back to it soon.

[jailln]

Hi @valentinMachado , thanks for your review and sorry for the late response.

What could I do to check if the solution is working ?

You can use a 3D Tiles dataset with bounding volumes of region type to test if it works. The only open source one and available online that I know is this one.

[mgermerie]

Except from the conflict that should be easy to fix it's all good for me. Thanks !