extorch.vision.transforms.segmentation

SegCompose	Transform compose for segmentation.
SegRandomHorizontalFlip	Horizontally flip the given image and label randomly with a given probability (Link).
SegNormalize	Normalization for segmentation, where the normalization is only applied on the input image.
SegCenterCrop	Crops the given image at the center.
SegRandomCrop	Random cropping for segmentation.
SegResize	Resize for segmentation.
SegRandomResize	Random resize for segmentation.
SegPILToTensor	PIL to Tensor for segmentation.
SegConvertImageDtype	Convert image dtype for segmentation.

class extorch.vision.transforms.segmentation.SegCenterCrop(size: Union[int, List[int]])

Bases: torch.nn.modules.module.Module

Crops the given image at the center.

Parameters

size (Union[int, List[int]]) – Height and width of the crop box.

forward(image, target)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class extorch.vision.transforms.segmentation.SegCompose(transforms)

Bases: torchvision.transforms.transforms.Compose

Transform compose for segmentation.

class extorch.vision.transforms.segmentation.SegConvertImageDtype(dtype)

Bases: torch.nn.modules.module.Module

Convert image dtype for segmentation.

forward(image, target)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class extorch.vision.transforms.segmentation.SegNormalize(mean: List[float], std: List[float])

Bases: torch.nn.modules.module.Module

Normalization for segmentation, where the normalization is only applied on the input image.

Parameters

• mean (List[float]) – List of means for each channel.

• std (List[float]) – List of standard deviations for each channel.

forward(image, target)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class extorch.vision.transforms.segmentation.SegPILToTensor

Bases: torch.nn.modules.module.Module

PIL to Tensor for segmentation.

forward(image, target)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class extorch.vision.transforms.segmentation.SegRandomCrop(size: Union[int, Tuple[int, int]])

Bases: torch.nn.modules.module.Module

Random cropping for segmentation. The Cropping is applied on the image and target at the same time.

Parameters

size (Union[int, Tuple[int, int]]) – Desired output size of the crop.

forward(image, target)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class extorch.vision.transforms.segmentation.SegRandomHorizontalFlip(p: float = 0.5)

Bases: torch.nn.modules.module.Module

Horizontally flip the given image and label randomly with a given probability (Link).

If the image and label are torch Tensor, they are expected to have […, H, W] shape, where … means an arbitrary number of leading dimensions.

Parameters

p (float) – probability of the image and label being flipped. Default: 0.5.

forward(image, target)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class extorch.vision.transforms.segmentation.SegRandomResize(min_size: Union[int, Tuple[int, int]], max_size: Optional[Union[T, Tuple[T, T]]] = None)

Bases: torch.nn.modules.module.Module

Random resize for segmentation.

Parameters

• min_size (Union[int, Tuple[int, int]]) – Desired minimum output size.

• max_size (Optional[Union[int, Tuple[int, int]]]) – Desired maximum output size. Default: None.

forward(image, target)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class extorch.vision.transforms.segmentation.SegResize(size: Union[int, Tuple[int, int]])

Bases: torch.nn.modules.module.Module

Resize for segmentation.

Parameters

size (Union[int, Tuple[int, int]]) – Desired output size.

forward(image, target)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool