mmtracking icon indicating copy to clipboard operation
mmtracking copied to clipboard

Published 20 hours ago • verified publisher icon open-mmlab

Implementation of interpolate_tracks and how to speed them up.

Open pvmilk opened this issue 2 years ago • 0 comments

Discussed in https://github.com/open-mmlab/mmtracking/discussions/907

Originally posted by pvmilk September 6, 2023 Does anyone have brief explanation of how this is implemented?

https://github.com/open-mmlab/mmtracking/blob/e79491ec8f0b8c86fda947fbaaa824c66ab2a991/mmtrack/core/track/interpolation.py#L5C16-L5C16

I would like to find out whether it is possible to speed up the calculation using parallelism. I am currently using it and found out that the calculation time seems to be exponential to the number of frames, e.g.

  • 25k frames would use ~70s
  • 75k frames would use ~700s

I am not sure whether this is proportionate to the number of frames OR the number of track_id.

pvmilk avatar Sep 05 '23 19:09 pvmilk