There are two parts in eval_bop19_pose.py: calculating errors and scoring. I batch the scoring part with 10 workers, and add this option in README.
This batching makes the run-time around 2x faster (of course without changing the score) on LM-O and TUD-L evaluation:
LM-O: 38.5 seconds instead of 73.5 seconds
TUD-L: 26.1 seconds instead of 59.2 seconds
I will batch the calculating errors part too (which is the only remaining bottleneck).
There are two parts in eval_bop19_pose.py: calculating errors and scoring. I batch the scoring part with 10 workers, and add this option in README.
This batching makes the run-time around 2x faster (of course without changing the score) on LM-O and TUD-L evaluation: LM-O: 38.5 seconds instead of 73.5 seconds TUD-L: 26.1 seconds instead of 59.2 seconds
I will batch the calculating errors part too (which is the only remaining bottleneck).
@thodan @ylabbe Please review and approve it!