print(len(dataset), "images in dataset")
# Plot of one image for comparision
random_idx = np.random.choice(range(len(dataset)))
plt.figure()
plt.subplot(1, 2, 1)
plt.imshow(dataset[random_idx][2], cmap='coolwarm')
plt.title("Noisy image")
plt.subplot(1, 2, 2)
plt.imshow(dataset[random_idx][5], cmap='coolwarm')
plt.title("Original image")
np.random.shuffle(dataset)
# Groundtruth: original clean image without edges
groundtruth = [(dataset[ix][0], dataset[ix][1], cut_edges_output(dataset[ix][2]), dataset[ix][3], dataset[ix][4],
cut_edges_output(dataset[ix][5])) for ix in range(len(dataset))]
# Input: downsampled groundtruth (antialiased)
downsampled = [(dataset[i][0], dataset[i][1], downsample2(groundtruth[i][2], 2), dataset[i][3], dataset[i][4],
groundtruth[i][5]) for i in range(len(dataset))]
# Reshaping of images
groundtruth = np.array(groundtruth)
downsampled = np.array(downsampled)
groundtruth_img = np.array([groundtruth_item[5].reshape(6, 10, 1) for groundtruth_item in groundtruth])
downsampled_img = np.array([downsampled_item[2].reshape(3, 5, 1) for downsampled_item in downsampled])
# Interpolation with tensorflow
interpolated_tens = tf.image.resize_images(downsampled_img, size=[6, 10], method=ResizeMethod.BICUBIC)
# Plot of one image for comparision
random_idx = np.random.choice(range(len(dataset)))
plt.figure()
plt.subplot(1, 2, 1)
plt.imshow(dataset[random_idx][2], cmap='coolwarm')
plt.title("Noisy image")
plt.subplot(1, 2, 2)
plt.imshow(dataset[random_idx][5], cmap='coolwarm')
plt.title("Original image")
np.random.shuffle(dataset)
# Groundtruth: original clean image without edges
groundtruth = [
(dataset[ix][0], dataset[ix][1], cut_edges_output(dataset[ix][2]),
dataset[ix][3], dataset[ix][4], cut_edges_output(dataset[ix][5]))
for ix in range(len(dataset))
]
# Input: downsampled groundtruth (antialiased)
downsampled = [
(dataset[i][0], dataset[i][1], downsample2(groundtruth[i][2], 2),
dataset[i][3], dataset[i][4], groundtruth[i][5])
for i in range(len(dataset))
]
# Splitting into train, validation and test sets and reshaping
groundtruth_train, groundtruth_val, groundtruth_test = np.split(
np.array(groundtruth),
[int(.8 * len(groundtruth)),
dataset_base[i][4], dataset_base[i][2]))
else:
dataset_test.append(
(dataset_base[i][0], dataset_base[i][1], im, dataset_base[i][3],
dataset_base[i][4], dataset_base[i][2]))
print("Length noisy dataset with modes removed: ", str(len(dataset_train)))
print("Length noisy dataset of remaining modes (test): ",
str(len(dataset_test)))
np.random.shuffle(dataset_train)
np.random.shuffle(dataset_test)
# Groundtruth: original clean image without edges
groundtruth_t = [(dataset_train[ix][0], dataset_train[ix][1],
cut_edges_output(dataset_train[ix][2]), dataset_train[ix][3],
dataset_train[ix][4], cut_edges_output(dataset_train[ix][5]))
for ix in range(len(dataset_train))]
# Input: downsampled groundtruth (antialiased)
downsampled_t = [(dataset_train[i][0], dataset_train[i][1],
downsample2(groundtruth_t[i][2], 2), dataset_train[i][3],
dataset_train[i][4], groundtruth_t[i][5])
for i in range(len(dataset_train))]
# Test sets
groundtruth_test = [(dataset_test[ix][0], dataset_test[ix][1],
cut_edges_output(dataset_test[ix][2]),
dataset_test[ix][3], dataset_test[ix][4],
cut_edges_output(dataset_test[ix][5]))
for ix in range(len(dataset_test))]
dataset_base[i][4], dataset_base[i][2]))
else:
dataset_test.append(
(dataset_base[i][0], dataset_base[i][1], im, dataset_base[i][3],
dataset_base[i][4], dataset_base[i][2]))
print("Length noisy dataset with modes removed: ", str(len(dataset_train)))
print("Length noisy dataset of remaining modes (test): ",
str(len(dataset_test)))
np.random.shuffle(dataset_train)
np.random.shuffle(dataset_test)
# Test sets
groundtruth_test = [(dataset_test[ix][0], dataset_test[ix][1],
cut_edges_output(dataset_test[ix][2]),
dataset_test[ix][3], dataset_test[ix][4],
cut_edges_output(dataset_test[ix][5]))
for ix in range(len(dataset_test))]
# Input: downsampled groundtruth (antialiased)
downsampled_test = [(dataset_test[i][0], dataset_test[i][1],
downsample2(groundtruth_test[i][2],
2), dataset_test[i][3], dataset_test[i][4],
groundtruth_test[i][5]) for i in range(len(dataset_test))]
groundtruth_test = np.array(groundtruth_test)
downsampled_test = np.array(downsampled_test)
groundtruth_test_img = np.array([
groundtruth_item[5].reshape(6, 10, 1)