Precomputed mesh chunk loading

I am able to generate multiresolution meshes and load them into neuroglancer. The meshes look good and the transitions seem okay at the lower resolutions, but towards the higher resolutions I notice that they don’t seem to load in an order that makes sense; in some cases things far off appear to load at higher resolutions before things up close. In some cases, the things up close never load at the high detail I would expect. See attached images, where the high res mesh has loaded for the part of the object that is farther away but never loads for the part that is closer (you can see the seam between chunks). Would this be expected where lower res chunks meet? Could this be because my higher resolution meshes take up too much memory? Or could this have to do with the fact I am setting grid_origin and vertex_offsets to [0,0,0], and calculate fragment_positions based on a given chunk’s overall position in the dataset?

[jbms]

I’m still investigating, but it appears that there is a problem with the octree that Neuroglancer constructs from the index file. I suspect it is a bug in Neuroglancer.

[davidackerman]

That did it, thanks so much! I naively was calculating Zorder using absolute fragment positions rather than their absolute positions divided by the chunk shape.