frarch.datasets.caltech101 module#

class frarch.datasets.caltech101.Caltech101(subset: str = 'train', transform: Optional[Callable] = None, target_transform: Optional[Callable] = None, root: Union[str, pathlib.Path] = './data/')[source]#

Bases: torch.utils.data.dataset.Dataset

Caltech 101 dataset object.

Data loader for the Caltech 101 dataset for object classification. The dataset can be obtained from https://data.caltech.edu/records/20086. The dataset is expected to have the following structure:

Expected dataset structure:

<root_folder>
├── class1
│   ├── image_xxxx.jpg
.   .
.   .
.   .
.   └── image_yyyy.jpg
└── classN
    ├── image_xxxx.jpg
    .
    .
    .
    └── image_yyyy.jpg
Parameters

  • subset (str) – “train” or “valid”. Subset to load. Defaults to “train”.

  • transform (Callable) – a callable object that takes an PIL.Image object and returns a modified PIL.Image object. Defaults to None, which won’t apply any transformation.

  • target_transform (Callable) – a callable object that the label data and returns modified label data. Defaults to None, which won’t apply any transformation.

  • root (Union[str, Path]) – root directory for the dataset. Defaults to ./data/.

References

Examples

Simple usage of the dataset class:

from frarch.datasets import Caltech101
from frarch.utils.data import create_dataloader
from torchvision.transforms import ToTensor
dataset = Caltech101("train", ToTensor, None, "./data/")
dataloader = create_dataloader(dataset)
for batch_idx, (batch, labels) in enumerate(dataloader):
    # process batch
get_number_classes() → int[source]#

Get number of target labels.

Returns

number of target labels.

Return type

int