"""Contains classes for image transformations specific for Big Earth Net dataset."""
import torch
from torchvision import transforms
from ..base import BaseTransforms
[docs]class ResizeToTensorNormalizeRGB(BaseTransforms):
"""
A class that applies resizing, tensor conversion, and normalization to RGB images.
"""
configurables = ["bands10_mean", "bands10_std"]
def __init__(self, *args, **kwargs):
"""
Initialize the class with the given mean and standard deviation for normalization.
:param bands10_mean: Mean values for the RGB bands
:type bands10_mean: list
:param bands10_std: Standard deviation values for the RGB bands
:type bands10_std: list
"""
super().__init__(self, *args, **kwargs)
self.bands10_mean = kwargs["bands10_mean"]
self.bands10_std = kwargs["bands10_std"]
def __call__(self, sample):
"""
Apply the transformation to the input sample.
:param sample: Input image
:type sample: PIL.Image.Image
:return: Transformed image
:rtype: torch.Tensor
"""
data_transforms = transforms.Compose([
transforms.ToTensor(), # transform the image from H x W x C to C x H x W
transforms.Resize((224, 224)),
transforms.Normalize(self.bands10_mean, self.bands10_std)
])
return data_transforms(sample)
[docs]class ToTensorResizeRandomCropFlipHV(BaseTransforms):
"""
A class that applies resizing, tensor conversion, random cropping, and random flipping to images.
"""
def __init__(self, *args, **kwargs):
super().__init__(self, *args, **kwargs)
def __call__(self, sample):
"""
Apply the transformation to the input sample.
:param sample: Input image
:type sample: PIL.Image.Image
:return: Transformed image
:rtype: torch.Tensor
"""
data_transforms = transforms.Compose([
transforms.ToTensor(), # transform the image from H x W x C to C x H x W
transforms.Resize((256, 256)),
transforms.RandomCrop(224),
transforms.RandomHorizontalFlip(),
transforms.RandomVerticalFlip(),
])
return data_transforms(sample)
[docs]class ToTensorResizeCenterCrop(BaseTransforms):
"""
A class that applies resizing, tensor conversion, and center cropping to images.
"""
def __init__(self, *args, **kwargs):
super().__init__(self, *args, **kwargs)
def __call__(self, sample):
"""
Apply the transformation to the input sample.
:param sample: Input image
:type sample: PIL.Image.Image
:return: Transformed image
:rtype: torch.Tensor
"""
data_transforms = transforms.Compose([
transforms.ToTensor(), # transform the image from H x W x C to C x H x W
transforms.Resize((256, 256)),
transforms.CenterCrop(224),
])
return data_transforms(sample)
[docs]class ToTensorResize(BaseTransforms):
"""
A class that applies resizing and tensor conversion to images.
"""
def __init__(self, *args, **kwargs):
super().__init__(self, *args, **kwargs)
def __call__(self, sample):
"""
Apply the transformation to the input sample.
:param sample: Input image
:type sample: PIL.Image.Image
:return: Transformed image
:rtype: torch.Tensor
"""
data_transforms = transforms.Compose([
transforms.ToTensor(), # transform the image from H x W x C to C x H x W
transforms.Resize((224, 224)),
])
return data_transforms(sample)
[docs]class NormalizeAllBands(BaseTransforms):
"""
A class that applies normalization to all bands of the input.
"""
configurables = ["bands10_mean", "bands10_std", "bands20_mean", "bands20_std"]
def __init__(self, *args, **kwargs):
"""
Initialize the class with the given mean and standard deviation for normalization.
:param bands10_mean: Mean values for the bands10
:type bands10_mean: list
:param bands10_std: Standard deviation values for the bands10
:type bands10_std: list
:param bands20_mean: Mean values for the bands20
:type bands20_mean: list
:param bands20_std: Standard deviation values for the bands20
:type bands20_std: list
"""
BaseTransforms.__init__(self, *args, **kwargs)
self.bands10_mean = kwargs["bands10_mean"]
self.bands10_std = kwargs["bands10_std"]
self.bands20_mean = kwargs["bands20_mean"]
self.bands20_std = kwargs["bands20_std"]
def __call__(self, input, target=None):
"""
Apply the transformation to the input bands.
:param input: List of input bands
:type sample: list
:return: Normalized bands
:rtype: list
"""
bands10, bands20, multihots = input
for t, m, s in zip(bands10, self.bands10_mean, self.bands10_std):
t.sub_(m).div_(s)
for t, m, s in zip(bands20, self.bands20_mean, self.bands20_std):
t.sub_(m).div_(s)
return bands10, bands20, multihots
[docs]class ToTensorAllBands(BaseTransforms):
"""
A class for converting all bands (list) to tensors.
"""
def __call__(self, input, target=None):
bands10, bands20, multihots = input
return torch.tensor(bands10).permute(2, 0, 1), torch.tensor(bands20).permute(2, 0, 1), multihots
