batch_size_pseudo=args.batch_size_pseudo,
state=0,
split=args.train_set,
input_sizes=input_sizes,
sets_id=args.sets_id,
mean=mean,
std=std,
keep_scale=keep_scale,
reverse_channels=reverse_channels)
after_loading()
net, optimizer = accelerator.prepare(net, optimizer)
time_now = time.time()
ratio = generate_class_balanced_pseudo_labels(
net=net,
device=device,
loader=unlabeled_loader,
input_size=input_sizes[2],
label_ratio=args.label_ratio,
num_classes=num_classes,
is_mixed_precision=args.mixed_precision)
print(ratio)
print('Pseudo labeling time: %.2fs' % (time.time() - time_now))
else:
labeled_loader, pseudo_labeled_loader, val_loader = init(
valtiny=args.valtiny,
no_aug=args.no_aug,
data_set=args.dataset,
batch_size_labeled=args.batch_size_labeled,
batch_size_pseudo=args.batch_size_pseudo,
state=1,
split=args.train_set,
input_sizes=input_sizes,
batch_size_labeled=args.batch_size_labeled,
batch_size_pseudo=args.batch_size_pseudo,
state=0,
split=args.train_set,
input_sizes=input_sizes,
sets_id=args.sets_id,
mean=mean,
std=std,
keep_scale=keep_scale,
reverse_channels=reverse_channels)
after_loading()
time_now = time.time()
generate_class_balanced_pseudo_labels(
net=net,
device=device,
loader=unlabeled_loader,
input_size=input_sizes[0],
label_ratio=args.label_ratio,
num_classes=num_classes)
print('Pseudo labeling time: %.2fs' % (time.time() - time_now))
else:
labeled_loader, pseudo_labeled_loader, val_loader = init(
valtiny=args.valtiny,
no_aug=args.no_aug,
data_set=args.dataset,
batch_size_labeled=args.batch_size_labeled,
batch_size_pseudo=args.batch_size_pseudo,
state=1,
split=args.train_set,
input_sizes=input_sizes,
sets_id=args.sets_id,