def generate_extra_data(X=None,
Y=None,
n=3000,
seed=None,
folder=io.DATA_FOLDER + 'train',
file_name=None,
rotate=True,
noise=True):
print "Creating {} new images".format(n)
if X is None and Y is None:
# automatically resizes images to (720, 1028)
X, _, _ = io.create_test_train_split(samples=-1,
auto_resize=True,
folder=folder)
X, Y = io.dataset_dict_to_array(X)
if seed:
np.random.seed(seed)
positions = np.random.randint(0, X.shape[0] - 1, n)
transformed = X[positions, :]
targets = Y[positions, :]
final = np.zeros_like(transformed)
final_resized_190_320 = np.zeros((n, 190, 320, 3))
final_resized_244_244 = np.zeros_like((n, 244, 244, 3))
final_resized_48_48 = np.zeros_like((n, 48, 48, 3))
# slight pertubations in the angle of the image
angles = np.random.uniform(-20, +20, n)
for i in range(n):
temp = transformed[i, :]
if rotate:
temp = imrotate(temp, angles[i], reshape=False)
if noise:
rand_noise = np.random.randint(0, 255, temp.shape)
keepers = np.random.binomial(1, 0.95, size=temp.shape)
temp = temp * keepers + rand_noise * (1 - keepers)
final[i, :] = temp
final_resized_190_320[i, :] = imresize(temp, size=(190, 320))
final_resized_244_244[i, :] = imresize(temp, size=(244, 244))
final_resized_48_48[i, :] = imresize(temp, size=(48, 48))
if i % 1000 == 0:
print "Created {} images.".format(i)
if file_name:
if not file_name.startswith("./data/"):
file_name = folder + file_name
np.save(file_name + '.npy', final)
np.save(file_name + '_190_320.npy', final_resized_190_320)
np.save(file_name + '_244_244.npy', final_resized_244_244)
np.save(file_name + '_48_48.npy', final_resized_48_48)
np.save(file_name + "_targets.npy", targets)
return final, targets
def load_imgs_from_paths(paths, auto_resize=True):
# shuffle paths so that when we do test/train splits
# we get shuffled distributions
paths = sampler(paths, len(paths))
if type(auto_resize) is tuple and len(auto_resize) == 2:
# resize to specified value
imgs = [imresize(read_image(path), size=auto_resize) for path in paths]
elif type(auto_resize) is bool and auto_resize:
# automatically resizes to image_utils.AUTO_RESIZE_DEFAULT
imgs = [imresize(read_image(path)) for path in paths]
else:
# no resize
imgs = [read_image(path) for path in paths]
return imgs
assert direction == 'a2b' or direction == 'b2a', 'Direction should be a2b or b2a!'
""" run """
frames = []
a_reals_ipt_ori = gif_frames(Image.open(gif_path))
size_ori = a_reals_ipt_ori.shape[0:3]
with tf.Session() as sess:
a_real = tf.placeholder(tf.float32,
shape=[None, crop_size, crop_size, crop_size, 3])
a2b = models.generator(a_real, direction)
# retore
saver = tf.train.Saver()
ckpt_path = utils.load_checkpoint('./checkpoints/' + dataset, sess, saver)
if ckpt_path is None:
raise Exception('No checkpoint!')
else:
print('Copy variables from % s' % ckpt_path)
a_real_ipt = np.zeros(
(1, len(a_real_ipt_ori), crop_size, crop_size, crop_size, 3))
a_real_ipt[0, ...] = im.imresize(a_real_ipt_ori,
[crop_size, crop_size, crop_size])
a2b_opt = sess.run(a2b, feed_dict={a_real: a_real_ipt})
a2b_opt_ori = im.imresize(a2b_opt[0, ..., 0].squeeze(), size_ori)
img_opt_ori = np.array([a_real_ipt_ori, a2b_opt_ori])
img_opt_ori = im.im2uint(im.immerge(img_opt_ori, 1, 1, 2))
writeGif(save_path, img_opt_ori, duration)
print('save in %s' % save_path)
shape = variable.get_shape()
print(shape)
print(len(shape))
variable_parameters = 1
for dim in shape:
print(dim)
variable_parameters *= dim.value
print(variable_parameters)
total_parameters += variable_parameters
print("\nTotal parameters:\n", total_parameters)
# start = time.time()
# Inference
for i in range(len(a_list)):
# Define shapes for images fed to the graph
a_feed = im.imresize(im.imread(a_list[i]), [crop_size, crop_size])
a_feed.shape = 1, crop_size, crop_size, 3
# Feed in images to the graph
a2b_result = sess.run(a2b, feed_dict={a_input: a_feed})
# Create and save the output image
a_img_opt = np.concatenate((a_feed, a2b_result), axis=0)
img_name = os.path.basename(a_list[i])
im.imwrite(im.immerge(a_img_opt, 1, 2), a_save_dir + '/' + img_name)
print('Save %s' % (a_save_dir + '/' + img_name))
# if i == 100:
# end = time.time()
# end2 = time.time()
# print("Time to process first 100 images:", end - start)
saver = tf.train.Saver()
ckpt_path = utils.load_checkpoint('./checkpoints/' + dataset, sess, saver)
if ckpt_path is None:
raise Exception('No checkpoint!')
else:
print('Copy variables from % s' % ckpt_path)
# test
a_list = glob('./datasets/' + dataset + '/testA/*.jpg')
b_list = glob('./datasets/' + dataset + '/testB/*.jpg')
a_save_dir = './test_predictions/' + dataset + '/testA/'
b_save_dir = './test_predictions/' + dataset + '/testB/'
utils.mkdir([a_save_dir, b_save_dir])
for i in range(len(a_list)):
a_real_ipt = im.imresize(im.imread(a_list[i]), [crop_size, crop_size])
a_real_ipt.shape = 1, crop_size, crop_size, 3
a2b_opt, a2b2a_opt = sess.run([a2b, a2b2a],
feed_dict={a_real: a_real_ipt})
a_img_opt = np.concatenate((a_real_ipt, a2b_opt, a2b2a_opt), axis=0)
img_name = os.path.basename(a_list[i])
im.imwrite(im.immerge(a_img_opt, 1, 3), a_save_dir + img_name)
print('Save %s' % (a_save_dir + img_name))
for i in range(len(b_list)):
b_real_ipt = im.imresize(im.imread(b_list[i]), [crop_size, crop_size])
b_real_ipt.shape = 1, crop_size, crop_size, 3
b2a_opt, b2a2b_opt = sess.run([b2a, b2a2b],
feed_dict={b_real: b_real_ipt})
b_img_opt = np.concatenate((b_real_ipt, b2a_opt, b2a2b_opt), axis=0)
assert direction == 'a2b' or direction == 'b2a', 'Direction should be a2b or b2a!'
""" run """
frames = []
a_reals_ipt_ori = gif_frames(Image.open(gif_path))
size_ori = a_reals_ipt_ori[0].shape[0:2]
with tf.Session() as sess:
a_real = tf.placeholder(tf.float32, shape=[None, crop_size, crop_size, 3])
a2b = models.generator(a_real, direction)
# retore
saver = tf.train.Saver()
ckpt_path = utils.load_checkpoint('./checkpoints/' + dataset, sess, saver)
if ckpt_path is None:
raise Exception('No checkpoint!')
else:
print('Copy variables from % s' % ckpt_path)
for a_real_ipt_ori in a_reals_ipt_ori:
a_real_ipt = im.imresize(a_real_ipt_ori, [crop_size, crop_size])
a_real_ipt.shape = 1, crop_size, crop_size, 3
a2b_opt = sess.run(a2b, feed_dict={a_real: a_real_ipt})
a2b_opt_ori = im.imresize(a2b_opt.squeeze(), size_ori)
img_opt_ori = np.array([a_real_ipt_ori, a2b_opt_ori])
img_opt_ori = im.im2uint(im.immerge(img_opt_ori, 1, 2))
frames.append(img_opt_ori)
writeGif(save_path, frames, duration)
print('save in %s' % save_path)
saver = tf.train.Saver()
ckpt_path = utils.load_checkpoint('./checkpoints/' + dataset, sess, saver)
if ckpt_path is None:
raise Exception('No checkpoint!')
else:
print('Copy variables from % s' % ckpt_path)
# test
a_list = glob('./datasets/' + dataset + '/testA/*.jpg')
b_list = glob('./datasets/' + dataset + '/testB/*.jpg')
a_save_dir = './test_predictions/' + dataset + '/testA/'
b_save_dir = './test_predictions/' + dataset + '/testB/'
utils.mkdir([a_save_dir, b_save_dir])
for i in range(len(a_list)):
a_real_ipt = im.imresize(im.imread(a_list[i]), [crop_size, crop_size])
a_real_ipt.shape = 1, crop_size, crop_size, 3
a2b_opt, a2b2a_opt = sess.run([a2b, a2b2a], feed_dict={a_real: a_real_ipt})
a_img_opt = np.concatenate((a_real_ipt, a2b_opt, a2b2a_opt), axis=0)
img_name = os.path.basename(a_list[i])
im.imwrite(im.immerge(a_img_opt, 1, 3), a_save_dir + img_name)
print('Save %s' % (a_save_dir + img_name))
for i in range(len(b_list)):
b_real_ipt = im.imresize(im.imread(b_list[i]), [crop_size, crop_size])
b_real_ipt.shape = 1, crop_size, crop_size, 3
b2a_opt, b2a2b_opt = sess.run([b2a, b2a2b], feed_dict={b_real: b_real_ipt})
b_img_opt = np.concatenate((b_real_ipt, b2a_opt, b2a2b_opt), axis=0)
img_name = os.path.basename(b_list[i])