def validate_model(self, images, epoch, step):
"""
:param images:
:param epoch:
:param step:
:return:
"""
fake_images, real_images, loss = self.generate_fake_samples(images)
print("Sample: loss: %.5f " % (loss))
merged_fake_images = merge(scale_back(fake_images),
[self.batch_size, 1])
merged_real_images = merge(scale_back(real_images),
[self.batch_size, 1])
merged_pair = np.concatenate([merged_fake_images, merged_real_images],
axis=1)
model_id, _ = self.get_model_id_and_dir()
model_sample_dir = os.path.join(self.sample_dir, model_id)
if not os.path.exists(model_sample_dir):
os.makedirs(model_sample_dir)
sample_img_path = os.path.join(model_sample_dir,
"sample_%04d_%06d.png" % (epoch, step))
misc.imsave(sample_img_path, merged_pair)
def infer(self, source_obj, model_dir, save_dir):
"""
:param source_obj:
:param model_dir:
:param save_dir:
:return:
"""
source_provider = InjectDataProvider(source_obj)
source_iter = source_provider.get_iter(self.batch_size)
tf.global_variables_initializer().run()
saver = tf.train.Saver(var_list=self.retrieve_generator_vars())
self.restore_model(saver, model_dir)
def save_imgs(imgs, count):
p = os.path.join(save_dir, "inferred_%04d.png" % count)
save_concat_images(imgs, img_path=p)
print("generated images saved at %s" % p)
count = 0
batch_buffer = list()
for source_imgs in source_iter:
fake_imgs = self.generate_fake_samples(source_imgs)[0]
merged_fake_images = merge(fake_imgs, [self.batch_size, 1])
batch_buffer.append(merged_fake_images)
if len(batch_buffer) == 10:
save_imgs(batch_buffer, count)
batch_buffer = list()
count += 1
if batch_buffer:
# last batch
save_imgs(batch_buffer, count)
def infer(self, source_obj, model_dir, save_dir):
"""
Inference this auto-encoder model.
:param source_obj:
:param model_dir:
:param save_dir:
:return:
"""
source_provider = InjectDataProvider(source_obj)
source_iter = source_provider.get_iter(self.batch_size)
tf.global_variables_initializer().run()
saver = tf.train.Saver(max_to_keep=100)
_, model_dir = self.get_model_id_and_dir()
self.restore_model(saver, model_dir)
def save_imgs(imgs, count):
p = os.path.join(save_dir, "inferred_%04d.png" % count)
save_concat_images(imgs, img_path=p)
print("generated images saved at %s" % p)
count = 0
batch_buffer = list()
code_list = None
for source_imgs in source_iter:
fake_imgs, real_imgs, loss, code = self.generate_fake_samples(
source_imgs)
if code_list is None:
code_list = code.copy()
else:
code_list = np.concatenate([code_list, code])
merged_fake_images = merge(fake_imgs, [self.batch_size, 1])
batch_buffer.append(merged_fake_images)
if len(batch_buffer) == 10:
save_imgs(batch_buffer, count)
batch_buffer = list()
count += 1
if batch_buffer:
# last batch
save_imgs(batch_buffer, count)
if code_list is not None:
code_list.dump(os.path.join(save_dir, "code.dat"))
print("code.dat dump successed!")
