peal.data.dataset_factory¶
Build the train / validation / test datasets described by a DataConfig.
This is the entry point every PEAL training script, adaptor and explainer uses to
turn a data yaml into PealDataset instances. It assembles the torchvision
transform pipelines (invariance augmentations for training, deterministic
resizing and cropping for evaluation), resolves the dataset class from
config.dataset_class or the input/output type, computes dataset statistics
for normalization when requested, and attaches the normalization and task
config to the returned datasets.
Functions
Instantiate the train, validation and test datasets of a data config. |