Conversion script for scannetppv2 writes enumerate() index as vertices_instance_id, mismatching source objectId

[Kevinpyx]

Summary: map_semantics_on_vertices uses the enumerate counter as the per-vertex instance id instead of the source segGroup["objectId"]. Combined with init-to-INVALID_ID=0, this causes (a) the first valid instance to alias background, and (b) any annotation indexed by the source objectId to point at the wrong object whenever the source segGroups have gaps or any entry was skipped due to label_ndx == INVALID_ID.

Fix: init with -1, write int(instance["objectId"]).

[Kevinpyx]

I realized the purpose of this enumerate() in the original conversion script is possibly to map the instance ids into a continuous sequence of ids starting from 1, which makes sense, but it stopped working when I want to use the labeled_mesh with other datasets that are annotated based on source objectId.