def test_vae(vae):
for part in ('train', 'val', 'test'):
input_images = dataset.fetch_smallbatch_from_celeba('CelebA',
part=part)
rec_images = vae.get_layer('decoder').predict(
vae.get_layer('encoder').predict(input_images)[0])
plot_image(img_renorm(input_images), img_renorm(rec_images))
def random(vae):
z_shape = K.int_shape(vae.get_layer('encoder').outputs[0])
z = K.random_normal(shape=(z_shape[1], ))
z = z.eval(session=tf.Session())
z = np.expand_dims(z, axis=0)
modified_image = vae.get_layer('decoder').predict(z)
plot_image(img_renorm(modified_image), img_renorm(modified_image))
def test_trans(translator, discriminator, image_file_name, target_gender):
image = read_image(image_file_name)
image = np.expand_dims(image, axis=0)
target_gender = np.expand_dims(target_gender, axis=0)
translated_img = translator.predict([image, target_gender])
r_src, _, r_cls = discriminator.predict(image)
g_src, _, g_cls = discriminator.predict(translated_img)
plot_image(img_renorm(image), img_renorm(translated_img))
print('input: ' + str(r_src) + " , " + str(r_cls) + ' - translated: ' +
str(g_src) + " , " + str(g_cls))
def trans_attribute(vae, image_file_name, attribute_vectors_file,
attr_trans_dic):
attribute_vectors = np.load(attribute_vectors_file).item()
image = read_image(image_file_name)
image = np.expand_dims(image, axis=0)
z = vae.get_layer('encoder').predict(image)[0]
for attr, trans in attr_trans_dic.items():
z[0] += (attribute_vectors[attr] * trans)
modified_image = vae.get_layer('decoder').predict(z)
plot_image(img_renorm(image), img_renorm(modified_image))
def compute_attribute_vector(vae,
attrs,
attribute_vectors_file,
batch_size=32):
sess = K.get_session()
encoder = vae.get_layer('encoder')
z_shape = K.int_shape(encoder.outputs[0])
pos_vectors = np.zeros((len(attrs), z_shape[1]), np.float32)
neg_vectors = np.zeros((len(attrs), z_shape[1]), np.float32)
pos_nums = np.zeros((len(attrs), 1), np.int32)
neg_nums = np.zeros((len(attrs), 1), np.int32)
data, total_num = dataset.load_full_celeba_with_labels(
'CelebA', batch_size, attrs)
iterator = data.make_one_shot_iterator()
next_element = iterator.get_next()
while True:
try:
images, labels = sess.run(next_element)
z = encoder.predict(images)[0]
for i in range(len(attrs)):
pos_idx = np.argwhere(labels[:, i] == 1)[:, 0]
neg_idx = np.argwhere(labels[:, i] == -1)[:, 0]
pos_vec = np.sum(z[pos_idx, :], 0)
neg_vec = np.sum(z[neg_idx, :], 0)
pos_nums[i][0] += len(pos_idx)
neg_nums[i][0] += len(neg_idx)
pos_vectors[i] += pos_vec
neg_vectors[i] += neg_vec
except tf.errors.OutOfRangeError:
break
pos_vectors /= pos_nums
neg_vectors /= neg_nums
attribute_vectors = {}
pos_images = vae.get_layer('decoder').predict(pos_vectors)
neg_images = vae.get_layer('decoder').predict(neg_vectors)
for i in range(len(attrs)):
attribute_vectors[attrs[i]] = pos_vectors[i] - neg_vectors[i]
# draw the attribute for debugging
print(attrs[i])
plot_image([img_renorm(pos_images[i])], [img_renorm(neg_images[i])])
np.save(attribute_vectors_file, attribute_vectors)
def test_gan(self, epoch):
for part in ('train', 'val', 'test'):
ds = dataset.load_celeba('CelebA',
batch_size,
part=part,
consumer='translator',
smallbatch=10)
element = ds.make_one_shot_iterator().get_next()
sess = K.get_session()
imgs, labels = sess.run(element)
labels = 1 - labels
print(labels)
rec_imgs = self.generator.predict([imgs, labels])
src_real, _, cls_real = self.discriminator.predict(imgs)
src_fake, _, cls_fake = self.discriminator.predict(rec_imgs)
for r, f, sr, cr, sf, cf in zip(imgs, rec_imgs, src_real, cls_real,
src_fake, cls_fake):
plot_images([img_renorm(r), img_renorm(f)])
print('real: ' + str(sr) + ' cls ' + str(cr) + ' fake: ' +
str(sf) + ' cls ' + str(cf))
def trans_attributes(vae, image_file_name, attribute_vectors_file, attrs):
attribute_vectors = np.load(attribute_vectors_file).item()
image = read_image('CelebA/img_align_celeba/' + image_file_name)
image = np.expand_dims(image, axis=0)
z = vae.get_layer('encoder').predict(image)[0]
z_v = [z[0] + attribute_vectors[attr] for attr in attrs]
z_v2 = [z[0] - attribute_vectors[attr] for attr in attrs]
z_v.extend(z_v2)
z_v = np.asarray(z_v, dtype=np.float32)
modified_images = vae.get_layer('decoder').predict(z_v)
images = np.append(modified_images, image, axis=0)
plot_images(img_renorm(images))
def merge_2(vae, image_file_name1, image_file_name2):
image1 = read_image(image_file_name1)
image1 = np.expand_dims(image1, axis=0)
z1 = vae.get_layer('encoder').predict(image1)[0]
image2 = read_image(image_file_name2)
image2 = np.expand_dims(image2, axis=0)
z2 = vae.get_layer('encoder').predict(image2)[0]
z_v = []
for i in range(9):
z = (z2[0] * i + z1[0] * (8 - i)) / 8
z_v.append(z)
z_v = np.asarray(z_v, dtype=np.float32)
images = vae.get_layer('decoder').predict(z_v)
plot_images(img_renorm(images))
def test_gan(self, epoch):
for part in ('train', 'val', 'test'):
images = dataset.fetch_smallbatch_from_celeba('CelebA', part=part)
rec_imgs = self.generator.predict(images)
plot_image(img_renorm(images), img_renorm(rec_imgs), epoch=epoch)