Implicit tiling spec

[ptrgags]

Direct link to 3DTILES_implicit_tiling spec

Opening this early because I wanted to refer to some of the availability buffer formulas in another issue.

This is the specification to describe the implicit tiling schema we merged into 3d-tiles-next last week.

I'll update this description once I have more of the text fleshed out.

[ptrgags]

Added some notes about problems with formulas

[ptrgags]

@sanjeetsuhag I just pushed the commits @ErixenCruz and I have been working on. There's still a couple sections and diagrams in progress, but you can review what we have so far

[sanjeetsuhag]

This diagram may potentially mislead someone into believing that binary trees are supported. We should use quadtrees.

[ptrgags]

Yeah, my only hesitation is that a quadtree diagram could get quite a bit wider (since we need to show the entire buffer).

As a group we decided to hold off for now and find a better solution later.

[ptrgags]

@sanjeetsuhag updated!

[ErixenCruz]

Should clarify binary tree is for visualization only since we only support quad and oct.

[ptrgags]

Found several small revisions, but nothing huge.

[ptrgags]

I feel like overall we could do better with examples if we had a more visual way of presenting at least one of them. Though as @ErixenCruz and I learned on Friday, putting together labeled examples can be time consuming (since determining morton order is necessary in some diagrams.

[ptrgags]

Note: we should be careful about subtree diagrams. subtree leaf nodes should point to the subtree below them, not a node directly.

[IanLilleyT]

Should

  "tileAvailability": {
    "constant": 0
  }

be explicitly forbidden?

Because if it were 0, there would be no point in having a subtree in the first place.

Another benefit of this constraint is if using tile metadata, it's guaranteed that a tile metadata buffer always exists.

[ptrgags]

@IanLilleyT @sanjeetsuhag @ErixenCruz @lilleyse

Changelist from this week's round of implicit tiling schema/spec updates:

• subtree files are binary now! the format is similar to GLB, but a bit more slimmed down (no JSON/BIN chunk headers)
• the description of buffers/bufferViews is now included in this spec since we're getting rid of 3DTILES_binary_buffers. This description includes the 8-byte alignment requirement.
• clarified some wording about the size of childSubtreeAvailability
• clarification: Levels are measured starting at 0 at the root implicit tile, i.e. the tile with the 3DTILES_implicit_tiling extension. This clarification is necessary since an implicit tileset may be part of a larger explicit tileset.
• 3DTILES_implicit_tiling.content is optional, and when present, subtree.contentAvailability is required. If content is not specified, then contentAvailability is disallowed.
• remove mimeType, it won't be useful until 2.0, and even then it's not a complete solution.
• Fixed some examples that still were putting the extension in the tileset rather than the root tile
• Added some details about availability that were present in the schema but not the spec.

What has NOT been changed (yet):

• the text of the spec still needs work. better diagrams and examples and a couple organization details. Full details in https://github.com/CesiumGS/3d-tiles-next/issues/119#issuecomment-746879326
• the example files need to be fixed, I need to transform the subtree.json files into binary files. I plan to do this early next week to avoid confusion.

Any questions/comments, please let me know.

[IanLilleyT]

• That is, if the deepest subtree. Should it be deepest subtree node?
• I assume L is a level local to the subtree, but it could be read as a global level.
• The equation N * (L + 1) should be N ^ (L + 1).
• No mention of morton encoding

I feel like this paragraph could be more clear if defined in terms of subtreeLevels

[lilleyse]

To keep the in-progress extensions in lockstep we're going to merge this PR into the 3d-tiles-next staging branch and work from there.