Computer_Vision/Chapter04/Image_augmentation.ipynb

%%capture
!pip install -U imgaug

import imgaug
print(imgaug.__version__)

0.4.0

import imgaug.augmenters as iaa

from torchvision import datasets
import torch
data_folder = '/content/' # This can be any directory you want to download FMNIST to
fmnist = datasets.FashionMNIST(data_folder, download=True, train=True)

/usr/local/lib/python3.7/dist-packages/torchvision/datasets/mnist.py:498: UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors. This means you can write to the underlying (supposedly non-writeable) NumPy array using the tensor. You may want to copy the array to protect its data or make it writeable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at  /pytorch/torch/csrc/utils/tensor_numpy.cpp:180.)
  return torch.from_numpy(parsed.astype(m[2], copy=False)).view(*s)

tr_images = fmnist.data
tr_targets = fmnist.targets

import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
from torch.utils.data import Dataset, DataLoader
import torch
import torch.nn as nn
device = 'cuda' if torch.cuda.is_available() else 'cpu'

def to_numpy(tensor):
    return tensor.cpu().detach().numpy()

plt.imshow(tr_images[0], cmap='gray')
plt.title('Original image')

Text(0.5, 1.0, 'Original image')

aug = iaa.Affine(scale=2)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])))
plt.title('Scaled image')

Text(0.5, 1.0, 'Scaled image')

aug = iaa.Affine(translate_px=10)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Translated image by 10 pixels (right and bottom)')

Text(0.5, 1.0, 'Translated image by 10 pixels (right and bottom)')

aug = iaa.Affine(translate_px={'x':10,'y':2})
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Translation of 10 pixels \nacross columns and 2 pixels over rows')

Text(0.5, 1.0, 'Translation of 10 pixels \nacross columns and 2 pixels over rows')

aug = iaa.Affine(rotate=30)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Rotation of image by 30 degrees')

Text(0.5, 1.0, 'Rotation of image by 30 degrees')

aug = iaa.Affine(rotate=-30)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Rotation of image by -30 degrees')

Text(0.5, 1.0, 'Rotation of image by -30 degrees')

aug = iaa.Affine(shear=30)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Shear of image by 30 degrees')

Text(0.5, 1.0, 'Shear of image by 30 degrees')

aug = iaa.Affine(shear=-30)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Shear of image by -30 degrees')

Text(0.5, 1.0, 'Shear of image by -30 degrees')

plt.figure(figsize=(20,20))
plt.subplot(161)
plt.imshow(tr_images[0], cmap='gray')
plt.title('Original image')
plt.subplot(162)
aug = iaa.Affine(scale=2, fit_output=True)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Scaled image')
plt.subplot(163)
aug = iaa.Affine(translate_px={'x':10,'y':2}, fit_output=True)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Translation of 10 pixels across \ncolumns and 2 pixels over rows')
plt.subplot(164)
aug = iaa.Affine(rotate=30, fit_output=True)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Rotation of image \nby 30 degrees')
plt.subplot(165)
aug = iaa.Affine(shear=30, fit_output=True)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Shear of image \nby 30 degrees')

Text(0.5, 1.0, 'Shear of image \nby 30 degrees')

plt.figure(figsize=(20,20))
plt.subplot(161)
plt.imshow(tr_images[0], cmap='gray')
plt.title('Original image')
plt.subplot(162)
aug = iaa.Affine(scale=2, fit_output=True)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Scaled image')
plt.subplot(163)
aug = iaa.Affine(translate_px={'x':10,'y':2}, fit_output=True, cval = 255)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Translation of 10 pixels across \ncolumns and 2 pixels over rows')
plt.subplot(164)
aug = iaa.Affine(rotate=30, fit_output=True)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Rotation of image \nby 30 degrees')
plt.subplot(165)
aug = iaa.Affine(shear=30, fit_output=True)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Shear of image \nby 30 degrees')

Text(0.5, 1.0, 'Shear of image \nby 30 degrees')

aug = iaa.Affine(rotate=30, fit_output=True, cval=255)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Rotation of image by 30 degrees')

Text(0.5, 1.0, 'Rotation of image by 30 degrees')

plt.figure(figsize=(20,20))
plt.subplot(161)
aug = iaa.Affine(rotate=30, fit_output=True, cval=0, mode='constant')
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Rotation of image by \n30 degrees with constant mode')
plt.subplot(162)
aug = iaa.Affine(rotate=30, fit_output=True, cval=0, mode='edge')
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Rotation of image by 30 degrees \n with edge mode')
plt.subplot(163)
aug = iaa.Affine(rotate=30, fit_output=True, cval=0, mode='symmetric')
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Rotation of image by \n30 degrees with symmetric mode')
plt.subplot(164)
aug = iaa.Affine(rotate=30, fit_output=True, cval=0, mode='reflect')
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Rotation of image by 30 degrees \n with reflect mode')
plt.subplot(165)
aug = iaa.Affine(rotate=30, fit_output=True, cval=0, mode='wrap')
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.title('Rotation of image by \n30 degrees with wrap mode')

Text(0.5, 1.0, 'Rotation of image by \n30 degrees with wrap mode')

plt.figure(figsize=(20,20))
plt.subplot(151)
aug = iaa.Affine(rotate=(-45,45), fit_output=True, cval=0, mode='constant')
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.subplot(152)
aug = iaa.Affine(rotate=(-45,45), fit_output=True, cval=0, mode='constant')
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.subplot(153)
aug = iaa.Affine(rotate=(-45,45), fit_output=True, cval=0, mode='constant')
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')
plt.subplot(154)
aug = iaa.Affine(rotate=(-45,45), fit_output=True, cval=0, mode='constant')
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')

<matplotlib.image.AxesImage at 0x7fc37927bf90>

aug = iaa.Multiply(1)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray')

<matplotlib.image.AxesImage at 0x7fc378d54d10>

aug = iaa.Multiply(0.5)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray',vmin = 0, vmax = 255)
plt.title('Pixels multiplied by 0.5')

Text(0.5, 1.0, 'Pixels multiplied by 0.5')

aug = iaa.LinearContrast(0.5)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray',vmin = 0, vmax = 255)
plt.title('Pixel contrast by 0.5')

Text(0.5, 1.0, 'Pixel contrast by 0.5')

aug = iaa.Dropout(p=0.2)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray',vmin = 0, vmax = 255)
plt.title('Random 20% pixel dropout')

Text(0.5, 1.0, 'Random 20% pixel dropout')

aug = iaa.SaltAndPepper(0.2)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray',vmin = 0, vmax = 255)
plt.title('Random 20% salt and pepper noise')

Text(0.5, 1.0, 'Random 20% salt and pepper noise')

plt.figure(figsize=(10,10))
plt.subplot(121)
aug = iaa.Dropout(p=0.2,)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray',vmin = 0, vmax = 255)
plt.title('Random 20% pixel dropout')
plt.subplot(122)
aug = iaa.SaltAndPepper(0.2,)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray',vmin = 0, vmax = 255)
plt.title('Random 20% salt and pepper noise')

Text(0.5, 1.0, 'Random 20% salt and pepper noise')

seq = iaa.Sequential([
      iaa.Dropout(p=0.2,),
    iaa.Affine(rotate=(-30,30))], random_order= True)

plt.imshow(seq.augment_image(to_numpy(tr_images[0])), cmap='gray',vmin = 0, vmax = 255)
plt.title('Image augmented using a \nrandom orderof the two augmentations')

Text(0.5, 1.0, 'Image augmented using a \nrandom orderof the two augmentations')

aug = iaa.GaussianBlur(sigma=1)
plt.imshow(aug.augment_image(to_numpy(tr_images[0])), cmap='gray',vmin = 0, vmax = 255)
plt.title('Gaussian blurring of image\n with a sigma of 1')

Text(0.5, 1.0, 'Gaussian blurring of image\n with a sigma of 1')