def prepare_inception_moments_eval_dataset(dataloader,
inception_model,
reduce_class,
splits,
logger,
device,
eval_dataset=False):
dataset_name = dataloader.dataset.dataset_name
if dataloader.dataset.train:
dataset_mode = 'train'
elif dataloader.dataset.train is False and dataset_name == "imagenet" or dataset_name == "tiny_imagenet":
dataset_mode = 'valid'
else:
dataset_mode = 'test'
inception_model.eval()
save_path = os.path.abspath(
os.path.join(
"./data", dataset_name + "_" + dataset_mode + "_" +
str(reduce_class) + '_inception_moments.npz'))
is_file = os.path.isfile(save_path)
is_score, is_std = None, None
if is_file is True:
mu = np.load(save_path)['mu']
sigma = np.load(save_path)['sigma']
else:
logger.info('Calculate moments of {} dataset'.format(dataset_mode))
mu, sigma = calculate_activation_statistics(
data_loader=dataloader,
generator=None,
discriminator=None,
inception_model=inception_model,
n_generate=None,
truncated_factor=None,
prior=None,
is_generate=False,
latent_op=False,
latent_op_step=None,
latent_op_alpha=None,
latent_op_beta=None,
device=device,
tqdm_disable=False)
logger.info('Saving calculated means and covariances to disk...')
np.savez(save_path, **{'mu': mu, 'sigma': sigma})
if eval_dataset is True:
logger.info(
'calculate inception score of {} dataset'.format(dataset_mode))
evaluator_instance = evaluator(inception_model, device=device)
is_score, is_std = evaluator_instance.eval_dataset(
dataloader, splits=len(dataloader.dataset) // 5000)
logger.info('Inception score={is_score}-Inception_std={is_std}'.format(
is_score=is_score, is_std=is_std))
return mu, sigma, is_score, is_std
def prepare_inception_moments(dataloader, eval_mode, generator,
inception_model, splits, run_name, logger,
device):
dataset_name = dataloader.dataset.dataset_name
inception_model.eval()
save_path = os.path.abspath(
os.path.join(
"./data",
dataset_name + "_" + eval_mode + '_' + 'inception_moments.npz'))
is_file = os.path.isfile(save_path)
if is_file:
mu = np.load(save_path)['mu']
sigma = np.load(save_path)['sigma']
else:
if device == 0:
logger.info('Calculate moments of {} dataset'.format(eval_mode))
mu, sigma = calculate_activation_statistics(
data_loader=dataloader,
generator=generator,
discriminator=None,
inception_model=inception_model,
n_generate=None,
truncated_factor=None,
prior=None,
is_generate=False,
latent_op=False,
latent_op_step=None,
latent_op_alpha=None,
latent_op_beta=None,
device=device,
tqdm_disable=False,
run_name=run_name)
if device == 0:
logger.info('Save calculated means and covariances to disk...')
np.savez(save_path, **{'mu': mu, 'sigma': sigma})
if is_file:
pass
else:
if device == 0:
logger.info(
'calculate inception score of {} dataset.'.format(eval_mode))
evaluator_instance = evaluator(inception_model, device=device)
is_score, is_std = evaluator_instance.eval_dataset(dataloader,
splits=splits)
if device == 0:
logger.info(
'Inception score={is_score}-Inception_std={is_std}'.format(
is_score=is_score, is_std=is_std))
return mu, sigma