Depth-anything-streaming colmap input?

Does the streaming version support colmap format input?

If not, that would be a good feature request, since if rough camera extrinsics are known, there's no need to recalculate them, and it will help keep all data if priors are known (cameras pointing at blank ceilings)

No files, tests, or entry points are named. Start by locating the streaming input path and checking how existing formats are parsed, then determine how COLMAP camera extrinsics and known priors would fit; done means the streaming version accepts the format while preserving the supplied camera information.

Written by the indexing model from the issue text.