def make_srcnn_rgb_dataset_based_on_cifar10():
print('making SRCNN-RGB dataset using CIFAR-10...')
(Y_train, _), (Y_test, _) = get_dataset_part(cifar10.load_data(),
train_part=0.2)
from image_handler import get_image, get_image_data, zoom_out_image, zoom_up_image
# X_train
print('making X_train list...')
X_train = []
for item in Y_train:
image_data = item.tolist()
image = get_image(image_data, mode='RGB')
zoomed_out_image = zoom_out_image(image, times=2)
zoomed_up_image = zoom_up_image(zoomed_out_image, times=2)
zoomed_up_image_data = get_image_data(zoomed_up_image)
X_train.append(zoomed_up_image_data)
# X_test
print('making X_test list...')
X_test = []
for item in Y_test:
image_data = item.tolist()
image = get_image(image_data, mode='RGB')
zoomed_out_image = zoom_out_image(image, times=2)
zoomed_up_image = zoom_up_image(zoomed_out_image, times=2)
zoomed_up_image_data = get_image_data(zoomed_up_image)
X_test.append(zoomed_up_image_data)
dtype = 'uint8'
dataset = (np.array(X_train, dtype=dtype), np.array(Y_train, dtype=dtype)), \
(np.array(X_test, dtype=dtype), np.array(Y_test, dtype=dtype))
print('saving dataset to srcnn-rgb-cifar10-dataset.npz...')
np.savez('datasets/srcnn-rgb-cifar10-dataset.npz', dataset)
def compare_images():
global first_image
global second_image
if first_image is None:
return
if second_image is None:
return
window = Toplevel()
window.title('Результаты измерений')
footer = Frame(window)
footer.pack(side=BOTTOM, expand=NO, fill=X)
ok_button = ToolbarButton(footer, text='ОК')
ok_button.pack(side=TOP, fill=Y, padx=10, pady=10)
ok_button.config(command=window.destroy)
ok_button.config(width=20)
true_image = image_handler.convert_to_rgb(first_image)
test_image = image_handler.convert_to_rgb(second_image)
true_image_data = image_handler.get_image_data(true_image)
test_image_data = image_handler.get_image_data(test_image)
true_image_data = np.array(true_image_data, dtype='uint8')
test_image_data = np.array(test_image_data, dtype='uint8')
# print(true_image_data.shape)
# print(test_image_data.shape)
psnr = compare_psnr(true_image_data, test_image_data)
ssim = compare_ssim(true_image_data, test_image_data, multichannel=True)
mse = compare_mse(true_image_data, test_image_data)
# print(true_image_data.shape)
# print(image_handler.get_image_head_data(first_image))
label_text = 'PSNR (peak signal-to-noise ratio) (Пиковое отношение сигнала к шуму): \n' + str(psnr) + ' dB' + '\n' + \
'SSIM (structural similarity) (Индекс структурного сходства) \n' + str(ssim) + '\n' + \
'MSE (mean squared error) (Среднеквадратичная ошибка)\n' + str(mse)
# TODO IFC (Information Fidelity Criterion), NQM (Noise Quality Measure),
# TODO PSNR (weighted peak signal-to-noise ratio),
# TODO MSSSIM (multiscale structure similarity index)
label = Label(window,
text=label_text,
font=COMPARE_IMAGES_RESULT_LABEL_FONT)
label.pack(side=TOP, fill=BOTH, expand=YES, padx=20, pady=20)
window.protocol('WM_DELETE_WINDOW', lambda: None)
window.grab_set()
window.focus_set()
window.wait_window()
def make_pasadena_dataset():
print('making PASADENA dataset...')
from image_handler import get_image_data, get_image, zoom_out_image, zoom_up_image
path = 'saved_images/Pasadena Dataset/'
filename_prefix = path + 'dcp_24'
size = 10
result = []
for i in range(size):
filename = filename_prefix + str(12 + i) + '.jpg'
image = Image.open(filename)
print('Image', filename.rpartition('/')[2], 'opened')
image_data = get_image_data(image)
result.append(image_data)
print('making Y_train and Y_test...')
train_size = 7
Y_train, Y_test = result[:train_size], result[train_size:]
# X_train
print('making X_train list...')
X_train = []
for item in Y_train:
image_data = item # .tolist()
image = get_image(image_data, mode='RGB')
zoomed_out_image = zoom_out_image(image, times=2)
zoomed_up_image = zoom_up_image(zoomed_out_image, times=2)
zoomed_up_image_data = get_image_data(zoomed_up_image)
X_train.append(zoomed_up_image_data)
# X_test
print('making X_test list...')
X_test = []
for item in Y_test:
image_data = item # .tolist()
image = get_image(image_data, mode='RGB')
zoomed_out_image = zoom_out_image(image, times=2)
zoomed_up_image = zoom_up_image(zoomed_out_image, times=2)
zoomed_up_image_data = get_image_data(zoomed_up_image)
X_test.append(zoomed_up_image_data)
dtype = 'uint8'
dataset = (np.array(X_train, dtype=dtype), np.array(Y_train, dtype=dtype)), \
(np.array(X_test, dtype=dtype), np.array(Y_test, dtype=dtype))
print('saving dataset to pasadena-dataset.npz...')
np.savez_compressed('datasets/pasadena-dataset.npz', dataset)
print('pasadena-dataset.npz saved')
def make_hundred_dataset():
print('making HUNDRED dataset...')
from image_handler import get_image_data, get_image, zoom_out_image, zoom_up_image
path = 'images/Hundred Dataset/images/'
size = 53
result = []
for i in range(size):
filename = path + str(1 + i) + '.png'
image = Image.open(filename)
print('Image', filename.rpartition('/')[2], 'opened')
image_data = get_image_data(image)
result.append(image_data)
print('making Y_train and Y_test...')
train_size = 40
Y_train, Y_test = result[:train_size], result[train_size:]
# X_train
print('making X_train list...')
X_train = []
for item in Y_train:
image_data = item # .tolist()
image = get_image(image_data, mode='RGB')
zoomed_out_image = zoom_out_image(image, times=2)
zoomed_up_image = zoom_up_image(zoomed_out_image, times=2)
zoomed_up_image_data = get_image_data(zoomed_up_image)
X_train.append(zoomed_up_image_data)
# X_test
print('making X_test list...')
X_test = []
for item in Y_test:
image_data = item # .tolist()
image = get_image(image_data, mode='RGB')
zoomed_out_image = zoom_out_image(image, times=2)
zoomed_up_image = zoom_up_image(zoomed_out_image, times=2)
zoomed_up_image_data = get_image_data(zoomed_up_image)
X_test.append(zoomed_up_image_data)
dtype = 'uint8'
dataset = (np.array(X_train, dtype=dtype), np.array(Y_train, dtype=dtype)), \
(np.array(X_test, dtype=dtype), np.array(Y_test, dtype=dtype))
print('saving dataset to hundred-dataset.npz...')
np.savez('datasets/hundred-dataset.npz', dataset) # _compressed
print('hundred-dataset.npz saved')
def handle_mnist():
from image_handler import get_image, get_image_data, zoom_out_image
from keras.datasets import mnist
(X_train, y_train), (X_test, y_test) = mnist.load_data()
dataset_size = 60000
dataset_X, dataset_Y = [], []
for i, X_train_item in enumerate(X_train[:dataset_size]):
image_data = X_train_item.tolist()
image = get_image(image_data, mode='L')
zoomed_out_image = zoom_out_image(image, times=2)
zoomed_out_image_data = get_image_data(zoomed_out_image)
dataset_X.append(zoomed_out_image_data)
dataset_Y.append(image_data)
print('image ' + str(i) + ', height : ' + str(len(image_data)),
'width : ' + str(len(image_data[0])),
sep=', ')
print('zoomed_out_image ' + str(i) + ', height : ' +
str(len(zoomed_out_image_data)),
'width : ' + str(len(zoomed_out_image_data[0])),
sep=', ')
# print('image_data :', image_data)
# print('zoomed_out_image_data :', zoomed_out_image_data)
# image.show()
# zoomed_out_image.show()
dataset = (dataset_X, dataset_Y)
import pickle
mnist_dataset_file = open('datasets/mnist-dataset.pkl', 'wb')
pickle.dump(dataset, mnist_dataset_file)
mnist_dataset_file.close()
def make_srcnn_dataset_based_on_mnist():
print('making SRCNN dataset using MNIST...')
from image_handler import get_image, get_image_data, zoom_out_image, zoom_up_image
(Y_train, _), (Y_test, _) = mnist.load_data() # 60000, 10000
# making X_train list
print('making X_train list...')
X_train = []
for item in Y_train:
image_data = item.tolist()
image = get_image(image_data, mode='L')
zoomed_out_image = zoom_out_image(image, times=2)
zoomed_up_image = zoom_up_image(zoomed_out_image, times=2)
zoomed_up_image_data = get_image_data(zoomed_up_image)
X_train.append(zoomed_up_image_data)
# making X_test list
print('making X_test list...')
X_test = []
for item in Y_test:
image_data = item.tolist()
image = get_image(image_data, mode='L')
zoomed_out_image = zoom_out_image(image, times=2)
zoomed_up_image = zoom_up_image(zoomed_out_image, times=2)
zoomed_up_image_data = get_image_data(zoomed_up_image)
X_test.append(zoomed_up_image_data)
# X_train, X_test = [], []
# for X, Y in [(X_train, Y_train), (X_test, Y_test)]:
# for item in Y:
# image_data = item.tolist()
# image = get_image(image_data, mode='L')
# zoomed_out_image = zoom_out_image(image, times=2)
# zoomed_up_image = zoom_up_image(zoomed_out_image, times=2)
# zoomed_up_image_data = get_image_data(zoomed_up_image)
# X.append(zoomed_up_image_data)
dtype = 'uint8'
dataset = (np.array(X_train, dtype=dtype), np.array(Y_train, dtype=dtype)), \
(np.array(X_test, dtype=dtype), np.array(Y_test, dtype=dtype))
print('saving dataset to srcnn-mnist-dataset.npz...')
np.savez('datasets/srcnn-mnist-dataset.npz', dataset)
