Simple example for a DA pipeline using Sequences

keras2.2.0_aug.py

import cv2
import numpy as np
from keras.preprocessing.image import ImageDataGenerator
from keras.utils import Sequence


class MySequence(Sequence):
    def __init__(self):
        self.path = '~/Images/cat.jpg'
        self.imgaug = ImageDataGenerator(rotation_range=20,
                                         rescale=1.,
                                         width_shift_range=10)

    def __len__(self):
        return 10

    def __getitem__(self, idx):
        X = np.array([cv2.resize(cv2.imread(self.path), (100, 100)) for _ in range(10)]).astype(np.float32)  # Fake batch of cats
        y = np.copy(X)
        for i in range(len(X)):
            params = self.imgaug.get_random_transform(X[i].shape)
            X[i] = self.imgaug.apply_transform(self.imgaug.standardize(X[i]), params)
            y[i] = self.imgaug.apply_transform(self.imgaug.standardize(y[i]), params)
        return X, y


if __name__ == '__main__':
    sequence = MySequence()
    for X, y in sequence:
        img = None
        for xi, yi in zip(X, y):
            im = np.concatenate((xi, yi), 0)
            if img is None:
                img = im
            else:
                img = np.concatenate((img, im), 1)
            cv2.imshow('wow', img.astype(np.uint8))
            cv2.waitKey(1000)

y[i] = self.imgaug.apply_transform(self.imgaug.standardize(y[i]), params)

Should you ever standardize y? I think it destroys the meaning of the target label.

For example, in my current application, y is a binary mask in the context of background/foreground segmentation. If you standardize y then its pixel values will be arbitrary floats rather than 1 or 0. And then metrics will always report zero accuracy because the model's prediction for each pixel value (1 or 0) always misses the new standardized values (arbitrary floats).

Is there any case where it makes sense to standardize y?