TinyImageNetDataModule#

class torch_uncertainty.datamodules.TinyImageNetDataModule(root, batch_size, eval_batch_size=None, eval_ood=False, eval_shift=False, shift_severity=1, val_split=None, num_tta=1, postprocess_set='val', train_transform=None, test_transform=None, ood_ds='svhn', interpolation='bilinear', basic_augment=True, rand_augment_opt=None, num_workers=1, pin_memory=True, persistent_workers=True)[source]#

DataModule for the Tiny-ImageNet dataset.

This datamodule uses Tiny-ImageNet as In-distribution dataset, OpenImage-O, ImageNet-0, SVHN or DTD as Out-of-distribution dataset and Tiny-ImageNet-C as shifted dataset.

Parameters:

  • root (str | Path) – Root directory of the datasets.

  • batch_size (int) – Number of samples per batch during training.

  • eval_batch_size (int | None) – Number of samples per batch during evaluation (val and test). Set to batch_size if None. Defaults to None.

  • eval_ood (bool) – Whether to evaluate out-of-distribution performance. Defaults to False.

  • eval_shift (bool) – Whether to evaluate on shifted data. Defaults to False.

  • num_tta (int) – Number of test-time augmentations (TTA). Defaults to 1 (no TTA).

  • shift_severity (int) – Severity of the shift. Defaults to 1.

  • val_split (float or Path) – Share of samples to use for validation or path to a yaml file containing a list of validation images ids. Defaults to 0.0.

  • postprocess_set (Literal['val', 'test']) – The post-hoc calibration dataset to use for the post-processing method. Defaults to val.

  • train_transform (Module | None) – Custom training transform. Defaults to None. If not provided, a default transform is used.

  • test_transform (Module | None) – Custom test transform. Defaults to None. If not provided, a default transform is used.

  • ood_ds (str) – Which out-of-distribution dataset to use. Defaults to "openimage-o".

  • test_alt – Which test set to use. Defaults to None.

  • procedure – Which procedure to use. Defaults to None.

  • train_size – Size of training images. Defaults to 224.

  • interpolation (str) – Interpolation method for the Resize Crops. Defaults to "bilinear".

  • basic_augment (bool) – Whether to apply base augmentations. Defaults to True.

  • rand_augment_opt (str | None) – Which RandAugment to use. Defaults to None.

  • num_workers (int) – Number of workers to use for data loading. Defaults to 1.

  • pin_memory (bool) – Whether to pin memory. Defaults to True.

  • persistent_workers (bool) – Whether to use persistent workers. Defaults to True.

test_dataloader()[source]#

Get test dataloaders for TinyImageNet.

Returns:

test set for in distribution data, OOD data, and/or TinyImageNetC data.

Return type:

list[DataLoader]