def test_polynomial_mmd_averages(self):
scores = kid_utils.polynomial_mmd_averages(codes_g=self.codes_g,
codes_r=self.codes_r,
n_subsets=4,
subset_size=1)
assert len(scores) == 4
assert type(scores[0]) == np.float64
def kid_score(num_samples,
netG,
dataset,
seed=0,
device=None,
num_subsets=10,
batch_size=50,
verbose=True,
feat_file=None,
log_dir='./log',
**kwargs):
"""
Computes KID score.
Args:
num_samples (int): The number of real and fake images to use for KID.
num_subsets (int): Number of subsets to compute average MMD.
netG (Module): Torch Module object representing the generator model.
device (str): Device identifier to use for computation.
seed (int): The random seed to use.
dataset (str/Dataset): The name of the dataset to load if known, or a custom Dataset object
batch_size (int): The batch size to feedforward for inference.
feat_file (str): The path to specific inception features for real images.
log_dir (str): Directory where features can be stored.
verbose (bool): If True, prints progress.
Returns:
tuple: Scalar mean and std of KID scores computed.
"""
start_time = time.time()
# Check inputs
if device is None:
device = torch.device('cuda:0' if torch.cuda.is_available() else "cpu")
if isinstance(dataset, str):
default_datasets = {
'cifar10',
'cifar100',
'stl10_48',
'imagenet_32',
'imagenet_128',
'celeba_64',
'celeba_128',
'lsun_bedroom',
'fake_data',
}
if dataset not in default_datasets:
raise ValueError('For default datasets, must be one of {}'.format(
default_datasets))
elif issubclass(type(dataset), torch.utils.data.Dataset):
if feat_file is None:
raise ValueError(
"feat_file to save/load from cannot be empty if using a custom dataset."
)
if not feat_file.endswith('.npz'):
feat_file = feat_file + '.npz'
else:
raise ValueError(
'dataset must be either a Dataset object or a string.')
# Make sure the random seeds are fixed
torch.manual_seed(seed)
random.seed(seed)
np.random.seed(seed)
# Directories
inception_path = os.path.join('./inception_model')
# Setup the inception graph
inception_utils.create_inception_graph(inception_path)
# Start producing features for real and fake images
# if device.index is not None:
# # Avoid unbounded memory usage
# gpu_options = tf.compat.v1.GPUOptions(allow_growth=True,
# per_process_gpu_memory_fraction=0.15,
# visible_device_list=str(device.index))
# config = tf.compat.v1.ConfigProto(gpu_options=gpu_options)
# else:
# config = tf.compat.v1.ConfigProto(device_count={'GPU': 0})
config = tf.compat.v1.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.2
config.gpu_options.allow_growth = True
with tf.compat.v1.Session(config=config) as sess:
sess.run(tf.compat.v1.global_variables_initializer())
real_feat = compute_real_dist_feat(num_samples=num_samples,
sess=sess,
dataset=dataset,
batch_size=batch_size,
verbose=verbose,
feat_file=feat_file,
log_dir=log_dir,
seed=seed,
**kwargs)
fake_feat = compute_gen_dist_feat(netG=netG,
num_samples=num_samples,
sess=sess,
device=device,
seed=seed,
batch_size=batch_size,
verbose=verbose)
# Compute the KID score
subset_size = num_samples // num_subsets
scores = kid_utils.polynomial_mmd_averages(real_feat,
fake_feat,
n_subsets=num_subsets,
subset_size=subset_size)
mmd_score, mmd_std = float(np.mean(scores)), float(np.std(scores))
print("INFO: KID: {:.4f} ± {:.4f} [Time Taken: {:.4f} secs]".format(
mmd_score, mmd_std,
time.time() - start_time))
return mmd_score, mmd_std
def kid_score(num_subsets,
subset_size,
netG,
device,
seed,
dataset_name,
batch_size=50,
verbose=True,
feat_file=None,
log_dir='./log'):
"""
Computes KID score.
Args:
num_subsets (int): Number of subsets to compute average MMD.
subset_size (int): Size of subset for computing MMD.
netG (Module): Torch Module object representing the generator model.
device (str): Device identifier to use for computation.
seed (int): The random seed to use.
dataset_name (str): The name of the dataset to load.
batch_size (int): The batch size to feedforward for inference.
feat_file (str): The path to specific inception features for real images.
log_dir (str): Directory where features can be stored.
verbose (bool): If True, prints progress.
Returns:
tuple: Scalar mean and std of KID scores computed.
"""
start_time = time.time()
# Make sure the random seeds are fixed
torch.manual_seed(seed)
random.seed(seed)
np.random.seed(seed)
# Directories
inception_path = os.path.join(log_dir, 'metrics', 'inception_model')
# Setup the inception graph
inception_utils.create_inception_graph(inception_path)
# Decide sample size
num_samples = int(num_subsets * subset_size)
# Start producing features for real and fake images
if device.index is not None:
# Avoid unbounded memory usage
gpu_options = tf.GPUOptions(allow_growth=True,
per_process_gpu_memory_fraction=0.15,
visible_device_list=str(device.index))
config = tf.compat.v1.ConfigProto(gpu_options=gpu_options)
else:
config = tf.compat.v1.ConfigProto(device_count={'GPU': 0})
with tf.compat.v1.Session(config=config) as sess:
sess.run(tf.compat.v1.global_variables_initializer())
real_feat = compute_real_dist_feat(num_samples=num_samples,
sess=sess,
dataset_name=dataset_name,
batch_size=batch_size,
verbose=verbose,
feat_file=feat_file,
log_dir=log_dir,
seed=seed)
fake_feat = compute_gen_dist_feat(netG=netG,
num_samples=num_samples,
sess=sess,
device=device,
seed=seed,
batch_size=batch_size,
verbose=verbose)
# Compute the KID score
scores = kid_utils.polynomial_mmd_averages(real_feat,
fake_feat,
n_subsets=num_subsets,
subset_size=subset_size)
mmd_score, mmd_std = float(np.mean(scores)), float(np.std(scores))
print("INFO: KID: {:.4f} ± {:.4f} [Time Taken: {:.4f} secs]".format(
mmd_score, mmd_std,
time.time() - start_time))
return mmd_score, mmd_std