blinklinmult.train

Training layer: configs, model, losses, metrics, and the Lightning loop.

• config — typed YAML configuration for a run.

• model — the eye-crop CNN plus LinMulT architecture.

• losses — masked binary losses for partially annotated supervision.

• metrics — masked detection metrics.

• module — the Lightning module.

• cli — the entry point make train-* calls.

Submodules

• blinklinmult.train.calibration
• blinklinmult.train.callbacks
• blinklinmult.train.cli
• blinklinmult.train.config
• blinklinmult.train.events
• blinklinmult.train.losses
• blinklinmult.train.lr_find
• blinklinmult.train.metrics
• blinklinmult.train.mlflow_utils
• blinklinmult.train.model
• blinklinmult.train.module
• blinklinmult.train.optim
• blinklinmult.train.phases