def save_checkpoint_and_samples(self, step, freq=1000):
if (np.mod(step, freq) == 0) and (self.d_counter == 0):
self.save_checkpoint(step)
samples = self.sess.run(self.sampler)
self._ensure_dirs('sample')
p = os.path.join(self.sample_dir, 'train_{:02d}.png'.format(step))
misc.save_images(samples[:64, :, :, :], [8, 8], p)
def test(model, data_loader, save_dir):
dataset = data_loader.load_data()
step = int(math.log2(dataset.dataset.get_resolution())) - 2
for i, data in enumerate(dataset):
generated = model.inference(data['style'], step=step)
visuals = OrderedDict([
('input_label', misc.tensor2im(data['style'][0])),
('synthesized_image', misc.tensor2im(generated.data[0]))
])
name = data['name'][0].split('/')[-1].split('.')[0]
misc.save_images(visuals, name, save_dir)
def get_samples(self, n=None, save=True, layers=[]):
if not (self.initialized_for_sampling or self.config.is_train):
print('[*] Loading from ' + self.checkpoint_dir + '...')
self.sess.run(tf.local_variables_initializer())
self.sess.run(tf.global_variables_initializer())
if self.load_checkpoint():
print(" [*] Load SUCCESS, model trained up to epoch %d" % \
self.sess.run(self.global_step))
else:
print(" [!] Load failed...")
return
if len(layers) > 0:
outputs = dict([(key + '_features', val)
for key, val in self.d_G_layers.items()])
if not (layers == 'all'):
keys = [sorted(list(outputs.keys()))[i] for i in layers]
outputs = dict([(key, outputs[key]) for key in keys])
else:
outputs = {}
outputs['samples'] = self.G
values = self._evaluate_tensors(outputs, n=n)
if not save:
if len(layers) > 0:
return values
return values['samples']
if not os.path.isdir(self.config.output_dir_of_test_samples):
os.mkdir(self.config.output_dir_of_test_samples)
for key, val in values.items():
if key == 'samples':
for idx in range(val.shape[0]):
print('Generating png to %s: %d / %d...' %
(self.config.output_dir_of_test_samples, idx,
val.shape[0]),
end='\r')
if self.config.model == 'mmd':
p = os.path.join(
self.config.output_dir_of_test_samples,
'MMD_{:08d}.png'.format(idx))
elif self.config.model == 'wgan_gp':
p = os.path.join(
self.config.output_dir_of_test_samples,
'WGAN-GP_{:08d}.png'.format(idx))
elif self.config.model == 'cramer':
p = os.path.join(
self.config.output_dir_of_test_samples,
'CRAMER_{:08d}.png'.format(idx))
misc.save_images(val[idx:idx + 1, :, :, :], [1, 1], p)