simet.providers.base.provider¶

simet.providers.base.provider ¶

Provider ¶

Provider(data_path)

Bases: ABC

Abstract data provider that builds a VisionDataset from a path + transform.

A Provider encapsulates how data is discovered and exposed as a torchvision.datasets.VisionDataset. It takes care of reading from data_path (e.g., folder layout, metadata) and applying a project-defined Transform to produce a dataset consumable by PyTorch DataLoaders.

Parameters:

Name	Type	Description	Default
`data_path`	`Path`	Root path where the underlying data resides (e.g., a folder tree).	required

Attributes:

Name	Type	Description
`data_path`	`Path`	The root path provided at construction.

Subclassing

Implement get_data(transform) to return a VisionDataset instance that will yield (image, target) pairs using the given Transform. The returned dataset should be ready to plug into a DataLoader.

Example

from torchvision.datasets import ImageFolder import torchvision.transforms as T

class FolderProvider(Provider): ... def get_data(self, transform: Transform) -> VisionDataset: ... tfm = transform.get_transform() ... return ImageFolder(self.data_path, transform=tfm) ... provider = FolderProvider(Path("data/train")) dataset = provider.get_data(transform=SomeTransform()) len(dataset) >= 0 True

Source code in simet/providers/base/provider.py

def __init__(self, data_path: Path) -> None:
    self.data_path = data_path

get_data `abstractmethod` ¶

get_data(transform)

Return a VisionDataset configured with the given transform.

Implementations should construct and return a dataset that reads from self.data_path and applies transform.get_transform() to each sample.

Parameters:

Name	Type	Description	Default
`transform`	`Transform`	Project transform wrapper used to build the underlying `torchvision.transforms.Compose`.	required

Returns:

Name	Type	Description
`VisionDataset`	`VisionDataset`	Dataset yielding `(sample, target)` pairs suitable for
	`VisionDataset`	consumption by a `torch.utils.data.DataLoader`.

Source code in simet/providers/base/provider.py

@abstractmethod
def get_data(self, transform: Transform) -> VisionDataset:
    """Return a `VisionDataset` configured with the given transform.

    Implementations should construct and return a dataset that reads from
    `self.data_path` and applies `transform.get_transform()` to each sample.

    Args:
        transform (Transform): Project transform wrapper used to build the
            underlying `torchvision.transforms.Compose`.

    Returns:
        VisionDataset: Dataset yielding `(sample, target)` pairs suitable for
        consumption by a `torch.utils.data.DataLoader`.
    """
    pass

simet.providers.base.provider¶