def _load_data():
# load CIFAR10 data
(target_data, _), (test_target_data, _) = cifar10.load_data()
# input image dimensions
# we assume data format "channels_last"
rows = target_data.shape[1]
cols = target_data.shape[2]
channels = target_data.shape[3]
# convert color train and test images to gray
source_data = other_utils.rgb2gray(target_data)
test_source_data = other_utils.rgb2gray(test_target_data)
# display color version of test images
imgs = test_target_data[:100]
title = 'CIFAR10 test color images (Ground Truth)'
img_shape = (rows, cols, channels)
filename = 'test_color.png'
other_utils.display_images(imgs,
img_shape=img_shape,
filename=filename,
title=title)
# display grayscale version of test images
imgs = test_source_data[:100]
title = 'CIFAR10 test gray images (Input)'
filename = 'test_gray.png'
other_utils.display_images(imgs,
img_shape=(rows, cols, 1),
filename=filename,
title=title)
# normalize output train and test color images
target_data = target_data.astype('float32') / 255
test_target_data = test_target_data.astype('float32') / 255
# normalize input train and test grayscale images
source_data = source_data.astype('float32') / 255
test_source_data = test_source_data.astype('float32') / 255
# reshape images to row x col x channels
# for CNN output/validation
target_data = target_data.reshape(target_data.shape[0], rows, cols,
channels)
test_target_data = test_target_data.reshape(test_target_data.shape[0],
rows, cols, channels)
# reshape images to row x col x channel for CNN input
source_data = source_data.reshape(source_data.shape[0], rows, cols, 1)
test_source_data = test_source_data.reshape(test_source_data.shape[0],
rows, cols, 1)
# source data, target data, test_source data
data = (source_data, target_data, test_source_data, test_target_data)
source_shape = (rows, cols, 1)
target_shape = (rows, cols, channels)
shapes = (source_shape, target_shape)
return data, shapes
def load_data():
# load CIFAR10 data
(x_train, _), (x_test, _) = cifar10.load_data()
# input image dimensions
# we assume data format "channels_last"
rows = x_train.shape[1]
cols = x_train.shape[2]
channels = x_train.shape[3]
# convert color train and test images to gray
x_train_gray = other_utils.rgb2gray(x_train)
x_test_gray = other_utils.rgb2gray(x_test)
# display color version of test images
imgs = x_test[:100]
title = 'Test color images (Ground Truth)'
img_shape = (rows, cols, channels)
filename = 'test_color.png'
other_utils.display_images(imgs,
img_shape=img_shape,
filename=filename,
title=title)
# display grayscale version of test images
imgs = x_test_gray[:100]
title = 'Test gray images (Input)'
filename = 'test_gray.png'
other_utils.display_images(imgs,
img_shape=(rows, cols, 1),
filename=filename,
title=title)
# normalize output train and test color images
x_train = x_train.astype('float32') / 255
x_test = x_test.astype('float32') / 255
# normalize input train and test grayscale images
x_train_gray = x_train_gray.astype('float32') / 255
x_test_gray = x_test_gray.astype('float32') / 255
# reshape images to row x col x channels
# for CNN output/validation
x_train = x_train.reshape(x_train.shape[0], rows, cols, channels)
x_test = x_test.reshape(x_test.shape[0], rows, cols, channels)
# reshape images to row x col x channel for CNN input
x_train_gray = x_train_gray.reshape(x_train_gray.shape[0], rows, cols, 1)
x_test_gray = x_test_gray.reshape(x_test_gray.shape[0], rows, cols, 1)
# source data, target data, test_source data
data = (x_train_gray, x_train, x_test_gray)
gray_shape = (rows, cols, 1)
color_shape = (rows, cols, channels)
# source shape, target shape
shapes = (gray_shape, color_shape)
return data, shapes
def load_cifar10():
(B_data, _), (test_B_data, _) = cifar10.load_data()
A_data = other_utils.rgb2gray(B_data)
test_A_data = other_utils.rgb2gray(test_B_data)
A_data = A_data[:, :, :, np.newaxis]
test_A_data = test_A_data[:, :, :, np.newaxis]
data = (A_data, B_data, test_A_data, test_B_data)
titles = ('CIFAR10 test_A_data images', 'CIFAR10 test_A_data images')
return other_utils.load_data(data, titles)
def load_data():
# load CIFAR10 data
(target_data, _), (test_target_data, _) = cifar10.load_data()
# input image dimensions
# we assume data format "channels_last"
rows = target_data.shape[1]
cols = target_data.shape[2]
channels = target_data.shape[3]
# convert color train and test images to gray
source_data = other_utils.rgb2gray(target_data)
test_source_data = other_utils.rgb2gray(test_target_data)
# reshape images to row x col x channel for CNN input
source_data = source_data.reshape(source_data.shape[0], rows, cols, 1)
test_source_data = test_source_data.reshape(test_source_data.shape[0],
rows, cols, 1)
# source data, target data, test_source data
data = (source_data, target_data, test_source_data, test_target_data)
filenames = ('cifar10_test_source.png', 'cifar10_test_target.png')
titles = ('CIFAR10 test source images', 'CIFAR10 test target images')
return other_utils.load_data(data, titles, filenames)
import numpy as np
from tensorflow.keras.datasets import cifar10
import matplotlib.pyplot as plt
import other_utils
import math
# load dataset
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
# sample cifar10 from train dataset
size = 1
side = int(math.sqrt(size))
indexes = np.random.randint(0, x_train.shape[0], size=size)
images = x_train[indexes]
gray_images = other_utils.rgb2gray(x_train[indexes])
# plot color cifar10
plt.figure(figsize=(side, side))
for i in range(len(indexes)):
plt.subplot(side, side, i + 1)
image = images[i]
plt.imshow(image)
plt.axis('off')
plt.savefig("cifar10-color-samples.png")
plt.show()
plt.close('all')
# plot gray cifar10
plt.figure(figsize=(side, side))