clarena.stl_datasets.imagenette

  1r"""
  2The submodule in `stl_datasets` for Imagenette dataset.
  3"""
  4
  5__all__ = ["Imagenette"]
  6
  7import logging
  8from typing import Callable
  9
 10import torch
 11from torch.utils.data import Dataset, random_split
 12from torchvision.datasets import Imagenette as ImagenetteRaw
 13from torchvision.transforms import transforms
 14
 15from clarena.stl_datasets.base import STLDatasetFromRaw
 16
 17# always get logger for built-in logging in each module
 18pylogger = logging.getLogger(__name__)
 19
 20
 21class Imagenette(STLDatasetFromRaw):
 22    r"""Imagenette dataset. The [Imagenette dataset](https://github.com/fastai/imagenette) is a subset of 10 easily classified classes from [Imagenet](https://www.image-net.org). It provides full sizes (as Imagenet), and resized 320x320 and 160x160. We support all of them in Permuted Imagenette."""
 23
 24    original_dataset_python_class: type[Dataset] = ImagenetteRaw
 25    r"""The original dataset class."""
 26
 27    def __init__(
 28        self,
 29        root: str,
 30        size: str,
 31        validation_percentage: float,
 32        batch_size: int = 1,
 33        num_workers: int = 0,
 34        custom_transforms: Callable | transforms.Compose | None = None,
 35        repeat_channels: int | None = None,
 36        to_tensor: bool = True,
 37        resize: tuple[int, int] | None = None,
 38    ) -> None:
 39        r"""
 40        **Args:**
 41        - **root** (`str`): the root directory where the original Imagenette data 'Imagenette/' live.
 42        - **size** (`str`): image size type. Supports "full" (default), "320px", and "160px".
 43        - **validation_percentage** (`float`): the percentage to randomly split some training data into validation data.
 44        - **batch_size** (`int`): The batch size in train, val, test dataloader.
 45        - **num_workers** (`int`): the number of workers for dataloaders.
 46        - **custom_transforms** (`transform` or `transforms.Compose` or `None`): the custom transforms to apply to ONLY TRAIN dataset. Can be a single transform, composed transforms or no transform. `ToTensor()`, normalize and so on are not included.
 47        - **repeat_channels** (`int` | `None`): the number of channels to repeat. Default is None, which means no repeat. If not None, it should be an integer.
 48        - **to_tensor** (`bool`): whether to include `ToTensor()` transform. Default is True.
 49        - **resize** (`tuple[int, int]` | `None` or list of them): the size to resize the images to. Default is None, which means no resize. If not None, it should be a tuple of two integers.
 50        """
 51        super().__init__(
 52            root=root,
 53            batch_size=batch_size,
 54            num_workers=num_workers,
 55            custom_transforms=custom_transforms,
 56            repeat_channels=repeat_channels,
 57            to_tensor=to_tensor,
 58            resize=resize,
 59        )
 60
 61        self.size: str = size
 62        r"""The size type of image."""
 63
 64        self.validation_percentage: float = validation_percentage
 65        r"""The percentage to randomly split some training data into validation data."""
 66
 67    def prepare_data(self) -> None:
 68        r"""Download the original Imagenette dataset if haven't."""
 69
 70        ImagenetteRaw(root=self.root, split="train", size=self.size, download=True)
 71        ImagenetteRaw(root=self.root, split="val", size=self.size, download=True)
 72
 73        pylogger.debug(
 74            "The original Imagenette dataset has been downloaded to %s.",
 75            self.root,
 76        )
 77
 78    def train_and_val_dataset(self) -> tuple[Dataset, Dataset]:
 79        """Get the training and validation dataset.
 80
 81        **Returns:**
 82        - **train_and_val_dataset** (`tuple[Dataset, Dataset]`): the train and validation dataset.
 83        """
 84        dataset_train_and_val = ImagenetteRaw(
 85            root=self.root,
 86            split="train",
 87            size=self.size,
 88            transform=self.train_and_val_transforms(),
 89            target_transform=self.target_transform(),
 90            download=False,
 91        )
 92
 93        return random_split(
 94            dataset_train_and_val,
 95            lengths=[1 - self.validation_percentage, self.validation_percentage],
 96            generator=torch.Generator().manual_seed(
 97                42
 98            ),  # this must be set fixed to make sure the datasets across experiments are the same. Don't handle it to global seed as it might vary across experiments
 99        )
100
101    def test_dataset(self) -> Dataset:
102        r"""Get the test dataset.
103
104        **Returns:**
105        - **test_dataset** (`Dataset`): the test dataset.
106        """
107        dataset_test = ImagenetteRaw(
108            root=self.root,
109            split="test",
110            size=self.size,
111            transform=self.test_transforms(),
112            target_transform=self.target_transform(),
113            download=False,
114        )
115
116        return dataset_test