ImageClassTrainer

class ImageClassTrainer(cfg)

Bases: Trainer

Trainer for dense 2D image batches (e.g. rasterized ring images).

Identical to Trainer except the train loader uses default_collate to stack per-event image into (B, C, H, W) and scalar labels/momenta into (B, 1), instead of the point-cloud collate that concatenates variable-length clouds along a single axis. Registered as ImageClassTrainer – select via train = dict(type="ImageClassTrainer").

build_train_loader()

Build the stateful image-classification training loader.

build_val_loader()

Build the optional image-classification validation loader.