SOTS
SOTS copied to clipboard
why scale and normalize embedding while training?
in SOTS/tree/MOT/CSTrack/lib/core/mot/base_trainer.py,why scale and normalize embedding while training?
after training,just use ps_id to compute cosine distance?