def evaluate(self, model_dir, config):
real_features = tf.contrib.gan.eval.run_inception(
images=tf.contrib.gan.eval.preprocess_image(self.real_images),
output_tensor="pool_3:0")
fake_features = tf.contrib.gan.eval.run_inception(
images=tf.contrib.gan.eval.preprocess_image(self.fake_images),
output_tensor="pool_3:0")
with tf.train.SingularMonitoredSession(scaffold=tf.train.Scaffold(
init_op=tf.global_variables_initializer(),
local_init_op=tf.group(tf.local_variables_initializer(),
tf.tables_initializer())),
checkpoint_dir=model_dir,
config=config) as session:
def generator():
while True:
try:
yield session.run([real_features, fake_features])
except tf.errors.OutOfRangeError:
break
frechet_inception_distance = metrics.frechet_inception_distance(
*map(np.concatenate, zip(*generator())))
tf.logging.info("frechet_inception_distance: {}".format(
frechet_inception_distance))
def compute_reconstruction_metrics(self):
epoch_id = self.get_last_epoch()
data = self.load_data(epoch_id)
recon = self.load_recon(epoch_id)
# TODO(nina): Rewrite mi and fid in pytorch
mutual_information = metrics.mutual_information(recon, data)
fid = metrics.frechet_inception_distance(recon, data)
data = torch.Tensor(data)
recon = torch.Tensor(recon)
bce = metrics.binary_cross_entropy(recon, data)
mse = metrics.mse(recon, data)
l1_norm = metrics.l1_norm(recon, data)
context = {
'title': 'Vaetree Report',
'bce': bce,
'mse': mse,
'l1_norm': l1_norm,
'mutual_information': mutual_information,
'fid': fid,
}
# Placeholder
return context
def validate(self):
def activation_generator():
self.generator.eval()
self.classifier.eval()
for real_images, _ in self.val_data_loader:
real_images = real_images.cuda(non_blocking=True)
latents = torch.randn(self.val_batch_size, self.latent_size, 1, 1)
latents = latents.cuda(non_blocking=True)
fake_images = self.generator(latents)
real_activations = self.classifier(real_images)
fake_activations = self.classifier(fake_images)
real_activations_list = [real_activations] * distributed.get_world_size()
fake_activations_list = [fake_activations] * distributed.get_world_size()
distributed.all_gather(real_activations_list, real_activations)
distributed.all_gather(fake_activations_list, fake_activations)
for real_activations, fake_activations in zip(real_activations_list, fake_activations_list):
yield real_activations, fake_activations
real_activations, fake_activations = map(torch.cat, zip(*activation_generator()))
frechet_inception_distance = metrics.frechet_inception_distance(real_activations.cpu().numpy(), fake_activations.cpu().numpy())
self.log_scalar('frechet_inception_distance', frechet_inception_distance)
def evaluate(self, model_dir, config, classifier, input_name,
output_names):
real_features, real_logits = tf.import_graph_def(
graph_def=classifier,
input_map={input_name: self.real_images},
return_elements=output_names)
fake_features, fake_logits = tf.import_graph_def(
graph_def=classifier,
input_map={input_name: self.fake_images},
return_elements=output_names)
with tf.train.SingularMonitoredSession(scaffold=tf.train.Scaffold(
init_op=tf.global_variables_initializer(),
local_init_op=tf.group(tf.local_variables_initializer(),
tf.tables_initializer())),
checkpoint_dir=model_dir,
config=config) as session:
def generator():
while not session.should_stop():
try:
yield session.run([real_features, fake_features])
except tf.errors.OutOfRangeError:
break
frechet_inception_distance = metrics.frechet_inception_distance(
*map(np.concatenate, zip(*generator())))
return dict(frechet_inception_distance=frechet_inception_distance)
def validate(self):
def create_activation_generator(data_loader):
def activation_generator():
self.inception.eval()
for real_images, _ in data_loader:
batch_size = real_images.size(0)
real_images = real_images.cuda(non_blocking=True)
latents = torch.randn(batch_size, self.latent_size, 1, 1)
latents = latents.cuda(non_blocking=True)
fake_images = self.generator(latents)
real_images = nn.functional.interpolate(real_images,
size=(299, 299),
mode="bilinear")
fake_images = nn.functional.interpolate(fake_images,
size=(299, 299),
mode="bilinear")
real_activations = self.inception(real_images)
fake_activations = self.inception(fake_images)
real_activations_list = [real_activations
] * self.world_size
fake_activations_list = [fake_activations
] * self.world_size
distributed.all_gather(real_activations_list,
real_activations)
distributed.all_gather(fake_activations_list,
fake_activations)
for real_activations, fake_activations in zip(
real_activations_list, fake_activations_list):
yield real_activations, fake_activations
return activation_generator
self.generator.eval()
self.discriminator.eval()
real_activations, fake_activations = map(
torch.cat,
zip(*create_activation_generator(self.val_data_loader)()))
frechet_inception_distance = metrics.frechet_inception_distance(
real_activations.cpu().numpy(),
fake_activations.cpu().numpy())
self.log_scalars(
{'frechet_inception_distance': frechet_inception_distance},
'validation')
summary_writer.add_scalars(main_tag="loss",
tag_scalar_dict=dict(
generator=generator_loss,
discriminator=discriminator_loss),
global_step=global_step)
global_step += 1
torch.save(generator.state_dict(),
f"{args.checkpoint_directory}/generator/epoch_{epoch}.pth")
torch.save(discriminator.state_dict(),
f"{args.checkpoint_directory}/discriminator/epoch_{epoch}.pth")
real_activations, fake_activations = map(
torch.cat, zip(*create_activation_generator(test_data_loader)()))
frechet_inception_distance = metrics.frechet_inception_distance(
real_activations.numpy(), fake_activations.numpy())
summary_writer.add_scalars(
main_tag="metrics",
tag_scalar_dict=dict(
frechet_inception_distance=frechet_inception_distance),
global_step=global_step)
summary_writer.export_scalars_to_json(f"events/scalars.json")
summary_writer.close()
print("----------------------------------------------------------------")
print(f"frechet_inception_distance: {frechet_inception_distance}")
print("----------------------------------------------------------------")