ITransform

Image transform that maps a tensor to a tensor.

Instance members

Compose

Compose multiple transforms into a single pipeline.

apply

apply transforms x

Apply transforms sequentially.

Parameters

• transforms : ITransform list
• x : Tensor

Returns Tensor

Normalize

Channel-wise normalization: $(x - \mu) / \sigma$. Input: $[C, H, W]$ or $[B, C, H, W]$.

Record fields

Instance members

Normalize module

imageNet

imageNet

ImageNet normalization preset.

Returns Normalize

Resize

Resize to the target spatial size using bilinear interpolation. Input: $[C, H, W]$ or $[B, C, H, W]$.

Record fields

Instance members

Resize module

create

create height width

Parameters

• height : int
• width : int

Returns Resize

RandomHorizontalFlip

Randomly flip the image horizontally with probability p. Input: $[C, H, W]$ or $[B, C, H, W]$.

Record fields

Instance members

RandomHorizontalFlip module

create

create p

Parameters

• p : float

Returns RandomHorizontalFlip

defaultFlip

defaultFlip

Returns RandomHorizontalFlip

RandomVerticalFlip

Randomly flip the image vertically with probability p. Input: $[C, H, W]$ or $[B, C, H, W]$.

Record fields

Instance members

RandomVerticalFlip module

create

create p

Parameters

• p : float

Returns RandomVerticalFlip

defaultFlip

defaultFlip

Returns RandomVerticalFlip

RandomCrop

Randomly crop the image to the given size. Input: $[C, H, W]$ or $[B, C, H, W]$.

Record fields

Instance members

RandomCrop module

create

create height width

Parameters

• height : int
• width : int

Returns RandomCrop

CenterCrop

Crop the center region of the image. Input: $[C, H, W]$ or $[B, C, H, W]$.

Record fields

Instance members

CenterCrop module

create

create height width

Parameters

• height : int
• width : int

Returns CenterCrop

ToGrayscale

Convert an RGB image to grayscale using ITU-R BT.601 weights. Input: $[3, H, W]$ or $[B, 3, H, W]$.

Record fields

Instance members

ToGrayscale module

create

create numOutputChannels

Parameters

• numOutputChannels : int

Returns ToGrayscale

single

single

Returns ToGrayscale

triple

triple

Returns ToGrayscale

ConvertImageDType

Convert image tensor dtype, scaling between [0, 255] int and [0.0, 1.0] float ranges. Input: any image tensor.

Record fields

Instance members

ConvertImageDType module

create

create dtype

Parameters

• dtype : ScalarType

Returns ConvertImageDType