osc.train

Main training methods.

Functions

batches_per_epoch_test	Compute number of batches in one test epoch.
batches_per_epoch_train	Compute number of batches in one training epoch.
batches_per_epoch_val	Compute number of batches in one validation epoch.
build_dataset_test	Build test dataset.
build_dataset_train	Build training dataset.
build_dataset_val	Build validation dataset.
build_dataset_vqa	Build VQA dataset for linear probing.
build_linear_probe	rtype Module
build_loss_fn_global	Build global loss function.
build_loss_fn_objects	Build object loss function.
build_model	Build model
build_optimizer	Build optimizer for training.
build_scheduler	Build learning rate scheduler for training.
extract_vqa_features	rtype Tuple[Tensor, Tensor]
get_viz_batch	Prepare a batch of images for visualization.
log_env_info
log_model_parameters	Log model parameters as a table.
main	rtype None
run_test_linear_probes
run_test_segmentation
run_train_epoch	Run one epoch of training.
run_train_val_viz_epochs	Run train, val, viz for a certain number of epochs.
run_val_epoch	Run one validation epoch.
run_viz	Run inference on a single batch of images and visualize everything!
update_cfg

Classes

ModelLoss

Function type signature for model loss functions.