Transforms

Transforms#

aitlas.transforms.big_earth_net module#

Contains classes for image transformations specific for Big Earth Net dataset.

class ResizeToTensorNormalizeRGB(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies resizing, tensor conversion, and normalization to RGB images.

Initialize the class with the given mean and standard deviation for normalization.

Parameters:

bands10_mean (list) – Mean values for the RGB bands
bands10_std (list) – Standard deviation values for the RGB bands

configurables = ['bands10_mean', 'bands10_std']#

class ToTensorResizeRandomCropFlipHV(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies resizing, tensor conversion, random cropping, and random flipping to images.

class ToTensorResizeCenterCrop(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies resizing, tensor conversion, and center cropping to images.

class ToTensorResize(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies resizing and tensor conversion to images.

class NormalizeAllBands(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies normalization to all bands of the input.

Initialize the class with the given mean and standard deviation for normalization.

Parameters:

bands10_mean (list) – Mean values for the bands10
bands10_std (list) – Standard deviation values for the bands10
bands20_mean (list) – Mean values for the bands20
bands20_std (list) – Standard deviation values for the bands20

configurables = ['bands10_mean', 'bands10_std', 'bands20_mean', 'bands20_std']#

class ToTensorAllBands(*args, **kwargs)[source]#

Bases: BaseTransforms

A class for converting all bands (list) to tensors.

aitlas.transforms.breizhcrops module#

Contains classes for image transformations specific for BreizhCrops dataset.

class SelectBands(*args, **kwargs)[source]#

Bases: BaseTransforms

A class used to select and process spectral bands from satellite data.

Parameters:: level (str) – satellite data level to be processed (“L1C” or “L2A”)

Note

This class requires a level argument at initialization. This should be one of the predefined satellite data levels (“L1C” or “L2A”).

Initialize the SelectBands class by setting the satellite data level.

configurables = ['level']#

aitlas.transforms.classification module#

Contains classes for image transformations for classification datasets.

class ResizeRandomCropFlipHVToTensor(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies resizing to (256,256), random cropping to size (224,224), random flipping, and tensor conversion to images.

class ResizeCenterCropFlipHVToTensor(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies resizing to (256,256), center cropping to size (224,224), random HV flipping, and tensor conversion to images.

class ResizeCenterCropToTensor(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies resizing to (256,256), center cropping to size (224,224), and tensor conversion to images.

class Resize1ToTensor(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies fixed resizing to (224,224) and tensor conversion to images.

class GrayToRGB(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that converts grayscale images to RGB format [height, width, channels].

class ConvertToRGBResizeCenterCropToTensor(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that converts an image to RGB format, applies resizing to size (256,256), center cropping to size (224,224), and tensor conversion.

class RandomFlipHVToTensor(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies random flipping and tensor conversion to images.

class ComplexTransform(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies complex transformations to images and tensor conversion.

The transformations include:

resizing to (256,256), random cropping to size (224,224),

random flipping (H and V) with probability 50%,

random brightness and constrast with probability 75%,

random blur (motion, median, gaussian, and noise) with probability 70%,

random distortion (optical, grid, elastic) with probability 70%,,

random CLAHE with probability 70%,

random HSV shift with probability 50%,

aitlas.transforms.joint_transforms module#

Contains joint transforms for images and label masks.

class FlipHVRandomRotate(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies flipping, random rotation, and shift-scale-rotation transformations to image and mask pairs.

class FlipHVToTensorV2(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies resizing, flipping, and tensor conversion to images with bounding boxes and labels.

class ResizeToTensorV2(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies resizing and tensor conversion to images with bounding boxes and labels.

class Resize(*args, **kwargs)[source]#

Bases: BaseTransforms

A class that applies resizing to images.

aitlas.transforms.object_detection module#

aitlas.transforms.segmentation module#

Classes and methods for image transformations for segmentation tasks. For semantic segmentation tasks the shape of the input is (N, 3, H, W); The shape of the output/mask is (N, num_classes, H, W), where N is the number of images

class MinMaxNormTranspose(*args, **kwargs)[source]#