Run-length encoded feature IDs

CesiumGS / 3d-tiles

Specification for streaming massive heterogeneous 3D geospatial datasets :earth_americas:

2.11k stars 467 forks source link

Run-length encoded feature IDs #366

Open nxddsnc opened 5 years ago

nxddsnc commented 5 years ago

I'm trying write a tool to batch mesh with same matereial in gltf file and convert it into b3dm file. But I don't understand the neccessary of putting batchId into vertex attributes. Since it will consume more memory than just put item id and it's start/end index in the batch table.

lilleyse commented 2 years ago

This is a good idea - essentially run-length encoding for feature IDs.

Couple follow up comments:

How well do feature IDs compress today?
Would feature IDs stay run-length encoded on the GPU? That would require gl_VertexID or equivalent to access that data (not available in WebGL 1.0). Or would they be unpacked to vertex attributes?

donmccurdy commented 2 years ago

Just a copy/paste of my comment in https://github.com/CesiumGS/3d-tiles/issues/572#issuecomment-969691195 but it's also relevant here –

If compression of the whole file delivers ~95% of the benefit, I think that's a really nice outcome because it can be handled in the HTTP layer, allowing arbitrary choice of compression method without affecting the specification. Might benefit from some attention to byte alignments of feature ID buffer views within the file.