benchmarks.hota.train_learned
module unitrack.benchmarks.hota.train_learnedTrain the learned MOTR-style appearance filter and write its checkpoint.
This is a one-time script that produces the small committed checkpoint
``benchmarks/hota/weights/learned_filter.safetensors`` consumed by
:func:`~unitrack.benchmarks.hota.tracker.build_learned_tracker`.
Members
function
- mainExtract train clips, train the learned filter, and save the checkpoint.
- save_checkpointWrite the flattened ``{propagator,fuser}`` state to a safetensors file.
- train_stepRun one contrastive multi-frame optimization step on a single clip.
constant
Source: unitrack/benchmarks/hota/train_learned.py:1