Deep learning notes - data expansion

Fixed shape ？ Convolutional neural network is usually a fixed input .

The effect is not certain . Many times I think it is useful because I think there will be pictures in the test set, which is similar to this effect .

So how to choose data enhancement ？ You can push back and forward , Think about the changes between the pictures in the deployment and test set and the pictures in the training set , To determine what kind of data enhancement is needed .

More diversity , Make generalization better . The purpose of the expansion is to make the training set more like the test set , It would be better if they were exactly the same ！:)

import torch
import torchvision
from torch import nn
from d2l import torch as d2l
import matplotlib.pyplot as plt


d2l.set_figsize()
img = d2l.Image.open('./img/cat1.jpg')
d2l.plt.imshow(img);


# aug: How to enlarge the picture 
def apply(img, aug, num_rows=2, num_cols=4, scale=1.5):
    Y = [aug(img) for _ in range(num_rows * num_cols)]
    d2l.show_images(Y, num_rows, num_cols, scale=scale)
    plt.show()


#  Flip image left and right 
# apply(img, torchvision.transforms.RandomHorizontalFlip())

#  Flip the image up and down 
# apply(img, torchvision.transforms.RandomVerticalFlip())

#  Random cutting 
shape_aug = torchvision.transforms.RandomResizedCrop(
    (200, 200), scale=(0.1, 1), ratio=(0.5, 2))

#  Randomly change the brightness of the image 
'''
apply(img, torchvision.transforms.ColorJitter(
    brightness=0.5, contrast=0, saturation=0, hue=0))
'''

#  Randomly change the hue of the image 
'''
apply(img, torchvision.transforms.ColorJitter(
    brightness=0, contrast=0, saturation=0, hue=0.5))
'''

#  Randomly change the brightness of the image 、 Contrast 、 Saturation and hue 
color_aug = torchvision.transforms.ColorJitter(
    brightness=0.5, contrast=0.5, saturation=0.5, hue=0.5)

#  Combined with a variety of image enhancement methods 
augs = torchvision.transforms.Compose([
    torchvision.transforms.RandomHorizontalFlip(), color_aug, shape_aug])
apply(img, augs)