extorch.vision.transforms.transforms

AdaptiveRandomCrop

Adaptively randomly crop images with uncertain sizes for a certain size.

AdaptiveCenterCrop

Adaptively center-crop images with uncertain sizes for a certain size.

Cutout

Cutout: Randomly mask out one or more patches from an image (Link).
class extorch.vision.transforms.transforms.AdaptiveCenterCrop(cropped_size: Union[int, Tuple[int, int]])[source]

Bases: torch.nn.modules.module.Module

Adaptively center-crop images with uncertain sizes for a certain size.

Parameters

cropped_size (Union[int, Tuple[int, int]]) – The Image size to be cropped out.

forward(img: torch.Tensor) → torch.Tensor[source]
Parameters

img (Tensor) – The image to be cropped.

Returns

The cropped image. For example, if the image has size [H, W] and the

cropped size if [h, w], size of output will be [H - h, W - w].

Return type

img (Tensor)

training: bool
class extorch.vision.transforms.transforms.AdaptiveRandomCrop(cropped_size: Union[int, Tuple[int, int]])[source]

Bases: torch.nn.modules.module.Module

Adaptively randomly crop images with uncertain sizes for a certain size.

Parameters

cropped_size (Union[int, Tuple[int, int]]) – The Image size to be cropped out.

forward(img: torch.Tensor) → torch.Tensor[source]
Parameters

img (Tensor) – The image to be cropped.

Returns

The cropped image. For example, if the image has size [H, W] and the

cropped size if [h, w], size of output will be [H - h, W - w].

Return type

img (Tensor)

training: bool
class extorch.vision.transforms.transforms.Cutout(length: int, n_holes: int = 1)[source]

Bases: torch.nn.modules.module.Module

Cutout: Randomly mask out one or more patches from an image (Link).

Parameters

  • length (int) – The length (in pixels) of each square patch.

  • image (Tensor) – Image of size (C, H, W).

  • n_holes (int) – Number of patches to cut out of each image. Default: 1.

Examples::
>>> image = torch.ones((3, 32, 32))
>>> Cutout_transform = Cutout(16, 1)
>>> image = Cutout_transform(image)  # Shape: [3, 32, 32]
forward(img: torch.Tensor) → torch.Tensor[source]
Parameters

img (Tensor) – Image of size (C, H, W).

Returns

Image with n_holes of dimension length x length cut out of it.

Return type

img (Tensor)

training: bool