A collection of information about Data Augmentation libraries for Image Processing.

torchvision

last updated on 10/10/2023

Repository Documentation Transformations
Supported languages:
  • Python
  • C++
  • Java
Supported ML libraries:
  • PyTorch
Transforming and augmenting images
Transforming and augmen...
Transforms scriptability
Transforms scriptability
Geometry
Geometry
Color
Color
Composition
Composition
Miscellaneous
Miscellaneous
Conversion
Conversion
Auto-Augmentation
Auto-Augmentation
Functional Transforms
Functional Transforms
torchvision
torchvisionText is not SVG - cannot display
Transformations Description Tags
1.

Data augmentation method based on AugMix.
Original Augmented
Original Augmented
Original Augmented 
out_img = T.AugMix(severity=3, mixture_width=3, chain_depth=-1,
         alpha=1.0, all_ops=True,
         interpolation=T.InterpolationMode.BILINEAR,
         fill=None)(img)
2.

Data augmentation method based on AutoAugment.
Original Augmented 
policy = T.AutoAugmentPolicy.IMAGENET
interpolation = T.InterpolationMode.NEAREST

out_img = T.AutoAugment(policy, interpolation, fill=None)(img)
3.

Crops the given image at the center.

 BASIC GEOMETRY
Original Augmented 
out_img = T.CenterCrop(size=800)(img)
Original Augmented 
out_img = T.CenterCrop(size=(800,1300))(img)
4.

Randomly changes the brightness, contrast, saturation and hue of an image.

 PHOTOMETRY
Original Augmented 
out_img = T.ColorJitter(brightness=(0,1), contrast=0, saturation=0,
                        hue=(0,0.5))(img)
Original Augmented 
out_img = T.ColorJitter(brightness=0, contrast=0.5, saturation=1,
                        hue=0)(img)
5.

Composes several transforms together.
6.

Crops the given image into four corners and the central crop.

 BASIC GEOMETRY
Original Augmented 
out_img = T.FiveCrop((400,300))(img)
7.

Blurs image with randomly chosen Gaussian blur.

 BLUR
Original Augmented 
out_img = T.GaussianBlur(19, sigma=(2.9, 3.0))(img)
8.

Converts image to grayscale.

 PHOTOMETRY
Original Augmented 
out_img = T.Grayscale(num_output_channels=1)(img)
9.

Pads the given image on all sides with the given pad value.

 BASIC GEOMETRY
Original Augmented 
out_img = T.Pad((5,20,50,30), fill=(200,10,100),
                padding_mode='constant')(img)
Original Augmented 
out_img = T.Pad(30, fill=0, padding_mode='edge')(img)
Original Augmented 
out_img = T.Pad(30, fill=0, padding_mode='reflect')(img)
10.

Data augmentation method based on RandAugment.
Original Augmented
Original Augmented 
out_img = T.RandAugment(num_ops=2, magnitude=9, num_magnitude_bins=31,
              interpolation=T.InterpolationMode.NEAREST, fill=None)(img)
11.

Adjusts the sharpness of the image randomly with a given probability.

 SHARPNESS
Original Augmented 
out_img = T.RandomAdjustSharpness(4, p=1)(img)
12.

Performs a random affine transformation of the image keeping center invariant.

 BASIC GEOMETRY
Original Augmented 
out_img = T.RandomAffine(45, translate=(0.1,0.5), scale=None, shear=None,
                         interpolation=T.InterpolationMode.NEAREST,
                         fill=(0,200,100), center=None)(img)
Original Augmented 
out_img = T.RandomAffine(0, translate=None, scale=(0.8,0.8),
                         shear=(10,10,20,20),
                         interpolation=T.InterpolationMode.NEAREST,
                         fill=0, center=None)(img)
13.

Applies randomly a list of transformations with a given probability.
14.

Autocontrasts the pixels of the given image randomly with a given probability.

 PHOTOMETRY
Original Augmented 
out_img = T.RandomAutocontrast(p=1)(img)

print(img.histogram() == out_img.histogram())
# >> False
Original Augmented 
out_img = T.RandomAutocontrast(p=1)(img)

print(img.histogram() == out_img.histogram())
# >> True
15.

Applies single transformation randomly picked from a list.
16.

Crops the given image at a random location.

 BASIC GEOMETRY
Original Augmented 
out_img = T.RandomCrop(500, padding=None, pad_if_needed=False,
                       fill=0, padding_mode='constant')(img)
17.

Equalizes the histogram of the given image randomly with a given probability.

 PHOTOMETRY
Original Augmented
Original Augmented 
print(img.histogram() == out_img.histogram())
18.

Randomly selects a rectangle region in a torch.Tensor image and erases its pixels. See here.

 DROPOUT
Original Augmented 
img = T.ToTensor()(img)
out_img = T.RandomErasing(p=1, scale=(0.02, 0.33), ratio=(0.3, 3.3),
                          value=0, inplace=False)(img)
out_img = T.ToPILImage()(out_img)
19.

Randomly converts image to grayscale with a probability of p.

 PHOTOMETRY
20.

Horizontally flips the given image randomly with a given probability.

 BASIC GEOMETRY
Original Augmented 
out_img = T.RandomHorizontalFlip(p=1)(img)
21.

Inverts the colors of the given image randomly with a given probability.

 PHOTOMETRY
Original Augmented 
out_img = T.RandomInvert(p=1)(img)
22.

Applies a list of transformations in a random order.
23.

Performs a random perspective transformation of the given image with a given probability.

 BASIC GEOMETRY
Original Augmented 
out_img = T.RandomPerspective(distortion_scale=0.5, p=1,
                              interpolation=T.InterpolationMode.BILINEAR,
                              fill=0)(img)
24.

Posterizes the image randomly with a given probability by reducing the number of bits for each color channel.

 STYLE FILTER
Original Augmented 
out_img = T.RandomPosterize(3, p=1)(img)
25.

Crops a random portion of image and resize it to a given size.

 BASIC GEOMETRY
Original Augmented 
out_img = T.RandomResizedCrop((300,400), scale=(0.08, 1.0),
                              ratio=(0.75, 1.33),
                              interpolation=T.InterpolationMode.BILINEAR,
                              antialias=None)(img)
26.

Rotates the image by angle.

 BASIC GEOMETRY
Original Augmented 
out_img = T.RandomRotation(90, interpolation=T.InterpolationMode.NEAREST,
                           expand=False, center=None, fill=0)(img)
Original Augmented 
out_img = T.RandomRotation(90, interpolation=T.InterpolationMode.NEAREST,
                           expand=True, center=None, fill=0)(img)
27.

Solarizes the image randomly with a given probability by inverting all pixel values above a threshold.

 PHOTOMETRY
Original Augmented 
out_img = T.RandomSolarize(180, p=1)(img)
28.

Vertically flips the given image randomly with a given probability.

 BASIC GEOMETRY
Original Augmented 
out_img = T.RandomVerticalFlip(p=1)(img)
29.

Resizes the input image to the given size.

 BASIC GEOMETRY
Original Augmented 
out_img = T.Resize(300, interpolation=T.InterpolationMode.BILINEAR, max_size=None, antialias=None)(img)
30.

Crops the given image into four corners and the central crop plus the flipped version of these (horizontal flipping is used by default).

 BASIC GEOMETRY
31.

Dataset-independent data-augmentation with TrivialAugment Wide.
Original Augmented
Original Augmented 
out_img = T.TrivialAugmentWide(num_magnitude_bins=31,
                     interpolation=T.InterpolationMode.NEAREST,
                     fill=None)(img)