helios.data.transforms

Attributes

TRANSFORM_REGISTRY

Global instance of the registry for transforms.

Classes

ToImageTensor

Convert an image (or list of images) to tensor(s).

Functions

create_transform(→ torch.nn.Module)

Create a transform of the given type.

Module Contents

helios.data.transforms.TRANSFORM_REGISTRY

Global instance of the registry for transforms.

Example

import helios.data.transforms as hldt

# This automatically registers your dataset.
@hldt.TRANSFORM_REGISTRY.register()
class MyTransform:
    ...

# Alternatively you can manually register a dataset like this:
hldt.TRANSFORM_REGISTRY.register(MyTransform)
helios.data.transforms.create_transform(type_name: str, *args: Any, **kwargs: Any) → torch.nn.Module

Create a transform of the given type.

This uses TRANSFORM_REGISTRY to look-up transform types, so ensure your transforms have been registered before using this function.

Parameters:

  • type_name – the type of the transform to create.

  • args – positional arguments to pass into the transform.

  • kwargs – keyword arguments to pass into the transform.

Returns:

The constructed transform.

class helios.data.transforms.ToImageTensor(dtype: torch.dtype = torch.float32, scale: bool = True)

Bases: torch.nn.Module

Convert an image (or list of images) to tensor(s).

An image is meant to be a tensor, ndarray, or PIL image. The shape expected to be either [H, W, C] or [C, H, W].

Parameters:

  • dtype – the output type of the tensors.

  • scale – if true, scale the values to the valid range. Defaults to true.

forward(img: numpy.typing.NDArray | list[numpy.typing.NDArray] | tuple[numpy.typing.NDArray, Ellipsis]) → torch.Tensor | list[torch.Tensor] | tuple[torch.Tensor, Ellipsis]

Convert the input image(s) into tensor(s).

The return type will match the type of the input. So, if the input is a single image, then the output will be a single tensor. If the input is a list or a tuple of images, the output will be a list or tuple of tensors.

Parameters:

img – image(s) to convert.

Returns:

The converted images.