def _map_augment_batch(*inputs):
image1, image2, image3, label = inputs
stddev = tf.random.uniform((1,), 0, 0.02, dtype=tf.float32)
flip_ud = tf.random.uniform([], 0, 1)
flip_rl = tf.random.uniform([], 0, 1)
blur = tf.random.uniform([], 0, 1)
brightness = tf.random.uniform((1,), -0.2, 0.2, dtype=tf.float32)
image1 = batch_augmentation(image1, stddev, flip_ud, flip_rl, brightness, blur)
image2 = batch_augmentation(image2, stddev, flip_ud, flip_rl, brightness, blur)
image3 = batch_augmentation(image3, stddev, flip_ud, flip_rl, brightness, blur)
return image1, image2, image3, label
def _map_augment_img(*inputs):
"""Data augmentation function to be applied"""
image, label = inputs
if augmentation_type == 'simple' or augmentation_type == 'batch':
image = tf.image.random_flip_left_right(image)
IMG_SIZE = image.shape[0] # CIFAR10: 32
# Add 4 pixels of padding
image = tf.image.resize_with_crop_or_pad(image, IMG_SIZE + 4, IMG_SIZE + 4)
# Random crop back to the original size
image = tf.image.random_crop(image, size=[IMG_SIZE, IMG_SIZE, 3])
elif augmentation_type == 'single':
# Special augmentation applied to each image individually to be comparable to metabyol
# First, simple aug, then special one
image = tf.image.random_flip_left_right(image)
IMG_SIZE = image.shape[0] # CIFAR10: 32
# Add 4 pixels of padding
image = tf.image.resize_with_crop_or_pad(image, IMG_SIZE + 4, IMG_SIZE + 4)
# Random crop back to the original size
image = tf.image.random_crop(image, size=[IMG_SIZE, IMG_SIZE, 3])
stddev = tf.random.uniform((1,), 0, 0.02, dtype=tf.float32)
flip_ud = tf.random.uniform([], 0, 1)
flip_rl = tf.random.uniform([], 0, 1)
blur = tf.random.uniform([], 0, 1)
brightness = tf.random.uniform((1,), -0.2, 0.2, dtype=tf.float32)
image = batch_augmentation(image, stddev, flip_ud, flip_rl, brightness, blur)
else:
assert 'augmentation type not supported!'
return image, label