# Register modules to checkpoint
checkpoint_io.register_modules(generator=generator, )
# Get model file
model_file = config['test']['model_file']
# Distributions
ydist = get_ydist(nlabels, device=device)
zdist = get_zdist('gauss', config['z_dist']['dim'], device=device)
# Test generator
generator_test = generator
# Evaluator
evaluator = Evaluator(generator_test,
zdist,
batch_size=config['test']['batch_size'],
device=device)
evaluator_single = Evaluator(generator_test,
zdist,
batch_size=config['test']['batch_size'],
device=device)
# Load checkpoint
load_dict = checkpoint_io.load(model_file)
it = load_dict.get('it', -1)
epoch_idx = load_dict.get('epoch_idx', -1)
# Pick a random but fixed seed
seed = torch.randint(0, 10000, (1, ))[0]
# Evaluation Loop
grid = np.reshape(grid, [-1, 2]).astype(np.float32)
grid = torch.from_numpy(grid).cuda()
grid_y = np.zeros([1000 * 1000]).astype(np.int64)
grid_y = torch.from_numpy(grid_y).cuda()
# Test generator
if config['training']['take_model_average']:
generator_test = copy.deepcopy(generator)
checkpoint_io.register_modules(generator_test=generator_test)
else:
generator_test = generator
# Evaluator
evaluator = Evaluator(generator_test,
zdist,
ydist,
batch_size=batch_size,
device=device)
# Train
tstart = t0 = time.time()
# Load checkpoint if it exists
try:
load_dict = checkpoint_io.load(model_file)
except FileNotFoundError:
it = epoch_idx = -1
else:
it = load_dict.get('it', -1)
epoch_idx = load_dict.get('epoch_idx', -1)
logger.load_stats('stats.p')
if config['test']['use_model_average']:
generator_test = copy.deepcopy(generator)
checkpoint_io.register_modules(generator_test=generator_test)
else:
generator_test = generator
# Distributions
ydist = get_ydist(nlabels, device=device)
zdist = get_zdist(config['z_dist']['type'],
config['z_dist']['dim'],
device=device)
# Evaluator
evaluator = Evaluator(generator_test,
zdist,
ydist,
batch_size=batch_size,
device=device)
# Load checkpoint if existant
load_dict = checkpoint_io.load(args.oldmodel)
it = load_dict.get('it', -1)
epoch_idx = load_dict.get('epoch_idx', -1)
# Inception score
if config['test']['compute_inception']:
print('Computing inception score...')
inception_mean, inception_std = evaluator.compute_inception_score()
print('Inception score: %.4f +- %.4f' % (inception_mean, inception_std))
# Samples
utils.save_images(x_real, path.join(out_dir, 'real.png'))
# Test generator
if config['training']['take_model_average']:
generator_test = copy.deepcopy(generator)
checkpoint_io.register_modules(generator_test=generator_test)
else:
generator_test = generator
# Evaluator
# NNN = 8000
x_real, _ = utils.get_nsamples(test_loader, NNN)
evaluator = Evaluator(generator_test,
zdist,
ydist,
batch_size=batch_size,
device=device,
fid_real_samples=x_real,
inception_nsamples=NNN,
fid_sample_size=NNN)
# Train
tstart = t0 = time.time()
it = -1
epoch_idx = -1
# Reinitialize model average if needed
if (config['training']['take_model_average']
and config['training']['model_average_reinit']):
update_average(generator_test, generator, 0.)
# Learning rate anneling
# Test generator
if config['training']['take_model_average']:
generator_test = copy.deepcopy(generator)
checkpoint_io.register_modules(generator_test=generator_test)
else:
generator_test = generator
# Evaluator
# evaluator = Evaluator(generator_test, zdist, ydist,
# batch_size=batch_size, device=device)
x_real_FID, _ = utils.get_nsamples(test_loader, NNN)
evaluator = Evaluator(generator_test,
zdist,
ydist,
batch_size=batch_size,
device=device,
fid_real_samples=x_real_FID,
inception_nsamples=NNN,
fid_sample_size=NNN)
it = -1
epoch_idx = -1
# Reinitialize model average if needed
if (config['training']['take_model_average']
and config['training']['model_average_reinit']):
update_average(generator_test, generator, 0.)
# Learning rate anneling
g_scheduler = build_lr_scheduler(g_optimizer, config, last_epoch=it)
d_scheduler = build_lr_scheduler(d_optimizer, config, last_epoch=it)
def perform_evaluation(run_name, image_type):
out_dir = os.path.join(os.getcwd(), '..', 'output', run_name)
checkpoint_dir = os.path.join(out_dir, 'chkpts')
checkpoints = sorted(glob.glob(os.path.join(checkpoint_dir, '*')))
evaluation_dict = {}
for point in checkpoints:
if not int(
point.split('/')[-1].split('_')[1].split('.')[0]) % 10000 == 0:
continue
iter_num = int(point.split('/')[-1].split('_')[1].split('.')[0])
model_file = point.split('/')[-1]
config = load_config('../configs/fr_default.yaml', None)
is_cuda = (torch.cuda.is_available())
checkpoint_io = CheckpointIO(checkpoint_dir=checkpoint_dir)
device = torch.device("cuda:0" if is_cuda else "cpu")
generator, discriminator = build_models(config)
# Put models on gpu if needed
generator = generator.to(device)
discriminator = discriminator.to(device)
# Use multiple GPUs if possible
generator = nn.DataParallel(generator)
discriminator = nn.DataParallel(discriminator)
generator_test_9 = copy.deepcopy(generator)
generator_test_99 = copy.deepcopy(generator)
generator_test_999 = copy.deepcopy(generator)
generator_test_9999 = copy.deepcopy(generator)
# Register modules to checkpoint
checkpoint_io.register_modules(
generator=generator,
generator_test_9=generator_test_9,
generator_test_99=generator_test_99,
generator_test_999=generator_test_999,
generator_test_9999=generator_test_9999,
discriminator=discriminator,
)
# Load checkpoint
load_dict = checkpoint_io.load(model_file)
# Distributions
ydist = get_ydist(config['data']['nlabels'], device=device)
zdist = get_zdist(config['z_dist']['type'],
config['z_dist']['dim'],
device=device)
z_sample = torch.Tensor(np.load('z_data.npy')).to(device)
#for name, model in zip(['0_', '09_', '099_', '0999_', '09999_'], [generator, generator_test_9, generator_test_99, generator_test_999, generator_test_9999]):
for name, model in zip(
['099_', '0999_', '09999_'],
[generator_test_99, generator_test_999, generator_test_9999]):
# Evaluator
evaluator = Evaluator(model, zdist, ydist, device=device)
x_sample = []
for i in range(10):
x = evaluator.create_samples(z_sample[i * 1000:(i + 1) * 1000])
x_sample.append(x)
x_sample = torch.cat(x_sample)
x_sample = x_sample / 2 + 0.5
if not os.path.exists('fake_data'):
os.makedirs('fake_data')
for i in range(10000):
torchvision.utils.save_image(x_sample[i, :, :, :],
'fake_data/{}.png'.format(i))
fid_score = calculate_fid_given_paths(
['fake_data', image_type + '_real'], 50, True, 2048)
print(iter_num, name, fid_score)
os.system("rm -rf " + "fake_data")
evaluation_dict[(iter_num, name[:-1])] = {'FID': fid_score}
if not os.path.exists('evaluation_data/' + run_name):
os.makedirs('evaluation_data/' + run_name)
pickle.dump(
evaluation_dict,
open('evaluation_data/' + run_name + '/eval_fid.p', 'wb'))
def main():
pp = pprint.PrettyPrinter(indent=1)
pp.pprint({
'data': config['data'],
'generator': config['generator'],
'discriminator': config['discriminator'],
'clusterer': config['clusterer'],
'training': config['training']
})
is_cuda = torch.cuda.is_available()
# Short hands
batch_size = config['training']['batch_size']
log_every = config['training']['log_every']
inception_every = config['training']['inception_every']
backup_every = config['training']['backup_every']
sample_nlabels = config['training']['sample_nlabels']
nlabels = config['data']['nlabels']
sample_nlabels = min(nlabels, sample_nlabels)
checkpoint_dir = path.join(out_dir, 'chkpts')
# Create missing directories
if not path.exists(out_dir):
os.makedirs(out_dir)
if not path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
# Logger
checkpoint_io = CheckpointIO(checkpoint_dir=checkpoint_dir)
device = torch.device("cuda:0" if is_cuda else "cpu")
train_dataset, _ = get_dataset(
name=config['data']['type'],
data_dir=config['data']['train_dir'],
size=config['data']['img_size'],
deterministic=config['data']['deterministic'])
train_loader = torch.utils.data.DataLoader(
train_dataset,
batch_size=batch_size,
num_workers=config['training']['nworkers'],
shuffle=True,
pin_memory=True,
sampler=None,
drop_last=True)
# Create models
generator, discriminator = build_models(config)
# Put models on gpu if needed
generator = generator.to(device)
discriminator = discriminator.to(device)
for name, module in discriminator.named_modules():
if isinstance(module, nn.Sigmoid):
print('Found sigmoid layer in discriminator; not compatible with BCE with logits')
exit()
g_optimizer, d_optimizer = build_optimizers(generator, discriminator, config)
devices = [int(x) for x in args.devices]
generator = nn.DataParallel(generator, device_ids=devices)
discriminator = nn.DataParallel(discriminator, device_ids=devices)
# Register modules to checkpoint
checkpoint_io.register_modules(generator=generator,
discriminator=discriminator,
g_optimizer=g_optimizer,
d_optimizer=d_optimizer)
# Logger
logger = Logger(log_dir=path.join(out_dir, 'logs'),
img_dir=path.join(out_dir, 'imgs'),
monitoring=config['training']['monitoring'],
monitoring_dir=path.join(out_dir, 'monitoring'))
# Distributions
ydist = get_ydist(nlabels, device=device)
zdist = get_zdist(config['z_dist']['type'], config['z_dist']['dim'], device=device)
ntest = config['training']['ntest']
x_test, y_test = utils.get_nsamples(train_loader, ntest)
x_cluster, y_cluster = utils.get_nsamples(train_loader, config['clusterer']['nimgs'])
x_test, y_test = x_test.to(device), y_test.to(device)
z_test = zdist.sample((ntest, ))
utils.save_images(x_test, path.join(out_dir, 'real.png'))
logger.add_imgs(x_test, 'gt', 0)
# Test generator
if config['training']['take_model_average']:
print('Taking model average')
bad_modules = [nn.BatchNorm1d, nn.BatchNorm2d, nn.BatchNorm3d]
for model in [generator, discriminator]:
for name, module in model.named_modules():
for bad_module in bad_modules:
if isinstance(module, bad_module):
print('Batch norm in discriminator not compatible with exponential moving average')
exit()
generator_test = copy.deepcopy(generator)
checkpoint_io.register_modules(generator_test=generator_test)
else:
generator_test = generator
clusterer = get_clusterer(config)(discriminator=discriminator,
x_cluster=x_cluster,
x_labels=y_cluster,
gt_nlabels=config['data']['nlabels'],
**config['clusterer']['kwargs'])
# Load checkpoint if it exists
it = utils.get_most_recent(checkpoint_dir, 'model') if args.model_it == -1 else args.model_it
it, epoch_idx, loaded_clusterer = checkpoint_io.load_models(it=it, load_samples='supervised' != config['clusterer']['name'])
if loaded_clusterer is None:
print('Initializing new clusterer. The first clustering can be quite slow.')
clusterer.recluster(discriminator=discriminator)
checkpoint_io.save_clusterer(clusterer, it=0)
np.savez(os.path.join(checkpoint_dir, 'cluster_samples.npz'), x=x_cluster)
else:
print('Using loaded clusterer')
clusterer = loaded_clusterer
# Evaluator
evaluator = Evaluator(
generator_test,
zdist,
ydist,
train_loader=train_loader,
clusterer=clusterer,
batch_size=batch_size,
device=device,
inception_nsamples=config['training']['inception_nsamples'])
# Trainer
trainer = Trainer(generator,
discriminator,
g_optimizer,
d_optimizer,
gan_type=config['training']['gan_type'],
reg_type=config['training']['reg_type'],
reg_param=config['training']['reg_param'])
# Training loop
print('Start training...')
while it < args.nepochs * len(train_loader):
epoch_idx += 1
for x_real, y in train_loader:
it += 1
x_real, y = x_real.to(device), y.to(device)
z = zdist.sample((batch_size, ))
y = clusterer.get_labels(x_real, y).to(device)
# Discriminator updates
dloss, reg = trainer.discriminator_trainstep(x_real, y, z)
logger.add('losses', 'discriminator', dloss, it=it)
logger.add('losses', 'regularizer', reg, it=it)
# Generators updates
gloss = trainer.generator_trainstep(y, z)
logger.add('losses', 'generator', gloss, it=it)
if config['training']['take_model_average']:
update_average(generator_test, generator, beta=config['training']['model_average_beta'])
# Print stats
if it % log_every == 0:
g_loss_last = logger.get_last('losses', 'generator')
d_loss_last = logger.get_last('losses', 'discriminator')
d_reg_last = logger.get_last('losses', 'regularizer')
print('[epoch %0d, it %4d] g_loss = %.4f, d_loss = %.4f, reg=%.4f'
% (epoch_idx, it, g_loss_last, d_loss_last, d_reg_last))
if it % config['training']['recluster_every'] == 0 and it > config['training']['burnin_time']:
# print cluster distribution for online methods
if it % 100 == 0 and config['training']['recluster_every'] <= 100:
print(f'[epoch {epoch_idx}, it {it}], distribution: {clusterer.get_label_distribution(x_real)}')
clusterer.recluster(discriminator=discriminator, x_batch=x_real)
# (i) Sample if necessary
if it % config['training']['sample_every'] == 0:
print('Creating samples...')
x = evaluator.create_samples(z_test, y_test)
x = evaluator.create_samples(z_test, clusterer.get_labels(x_test, y_test).to(device))
logger.add_imgs(x, 'all', it)
for y_inst in range(sample_nlabels):
x = evaluator.create_samples(z_test, y_inst)
logger.add_imgs(x, '%04d' % y_inst, it)
# (ii) Compute inception if necessary
if it % inception_every == 0 and it > 0:
print('PyTorch Inception score...')
inception_mean, inception_std = evaluator.compute_inception_score()
logger.add('metrics', 'pt_inception_mean', inception_mean, it=it)
logger.add('metrics', 'pt_inception_stddev', inception_std, it=it)
print(f'[epoch {epoch_idx}, it {it}] pt_inception_mean: {inception_mean}, pt_inception_stddev: {inception_std}')
# (iii) Backup if necessary
if it % backup_every == 0:
print('Saving backup...')
checkpoint_io.save('model_%08d.pt' % it, it=it)
checkpoint_io.save_clusterer(clusterer, int(it))
logger.save_stats('stats_%08d.p' % it)
if it > 0:
checkpoint_io.save('model.pt', it=it)
if config['test']['use_model_average']:
generator_test = copy.deepcopy(generator)
checkpoint_io.register_modules(generator_test=generator_test)
else:
generator_test = generator
# Distributions
ydist = get_ydist(nlabels, device=device)
zdist = get_zdist(config['z_dist']['type'],
config['z_dist']['dim'],
device=device)
# Evaluator
evaluator = Evaluator(generator_test,
zdist,
ydist,
batch_size=batch_size,
device=device)
print(generator.module.resnet_0_0.conv_0.weight[1, 1, :, :])
# Load checkpoint if existant
load_dict = checkpoint_io.load(model_file)
it = load_dict.get('it', -1)
epoch_idx = load_dict.get('epoch_idx', -1)
print(generator.module.resnet_0_0.conv_0.weight[1, 1, :, :])
TrainModeSave = DATA
torch.save(generator.module.state_dict(),
save_dir + TrainModeSave + 'Pre_generator')
utils.save_images(x_real, path.join(out_dir, 'real.png'))
# Test generator
if config['training']['take_model_average']:
generator_test = copy.deepcopy(generator)
checkpoint_io.register_modules(generator_test=generator_test)
else:
generator_test = generator
# Evaluator
if inception_every > 0 and compute_fid:
# This will also compute FID
# Load fid_samples (1024) many.
fid_real_samples, _ = utils.get_nsamples(train_loader, fid_sample_size)
evaluator = Evaluator(generator_test, zdist, ydist,
batch_size=batch_size, device=device,
fid_real_samples=fid_real_samples,
fid_sample_size=fid_sample_size)
else:
evaluator = Evaluator(generator_test, zdist, ydist,
batch_size=batch_size, device=device)
# Train
tstart = t0 = time.time()
it = epoch_idx = -1
# Load checkpoint if existant
it = checkpoint_io.load('model.pt')
if it != -1:
logger.load_stats('stats.p')
if adaptive_beta:
# Set reg_param to the last reg_param
reg_param = logger.stats['learning_rates']['beta_value'][-1]
path = "Final_cifar_nopid_sigmoid.1_"
for epoch_id in range(40, 80):
model_name = "/home/kunxu/Workspace/GAN_PID/output/" + path
config = load_config(os.path.join(model_name, "config.yaml"),
'configs/default.yaml')
generator, discriminator = build_models(config)
generator = torch.nn.DataParallel(generator)
zdist = get_zdist(config['z_dist']['type'],
config['z_dist']['dim'],
device=device)
ydist = get_ydist(1, device=device)
checkpoint_io = CheckpointIO(checkpoint_dir="./tmp")
checkpoint_io.register_modules(generator_test=generator)
evaluator = Evaluator(generator,
zdist,
ydist,
batch_size=100,
device=device)
ckptpath = os.path.join(model_name, "chkpts",
"model_{:08d}.pt".format(epoch_id * 10000 + 9999))
print(ckptpath)
load_dict = checkpoint_io.load(ckptpath)
img_list = []
for i in range(500):
ztest = zdist.sample((100, ))
x = evaluator.create_samples(ztest)
img_list.append(x.cpu().numpy())
img_list = np.concatenate(img_list, axis=0)
m, s = evaluation(img_list)
all_results.append([float(m), float(s)])