def train(train_source_iter: ForeverDataIterator, train_target_iter: ForeverDataIterator, model: ImageClassifier,
jmmd_loss: JointMultipleKernelMaximumMeanDiscrepancy, optimizer: SGD,
lr_scheduler: LambdaLR, epoch: int, args: argparse.Namespace):
batch_time = AverageMeter('Time', ':4.2f')
data_time = AverageMeter('Data', ':3.1f')
losses = AverageMeter('Loss', ':3.2f')
trans_losses = AverageMeter('Trans Loss', ':5.4f')
cls_accs = AverageMeter('Cls Acc', ':3.1f')
tgt_accs = AverageMeter('Tgt Acc', ':3.1f')
progress = ProgressMeter(
args.iters_per_epoch,
[batch_time, data_time, losses, trans_losses, cls_accs, tgt_accs],
prefix="Epoch: [{}]".format(epoch))
# switch to train mode
model.train()
jmmd_loss.train()
end = time.time()
for i in range(args.iters_per_epoch):
x_s, labels_s = next(train_source_iter)
x_t, labels_t = next(train_target_iter)
x_s = x_s.to(device)
x_t = x_t.to(device)
labels_s = labels_s.to(device)
labels_t = labels_t.to(device)
# measure data loading time
data_time.update(time.time() - end)
# compute output
x = torch.cat((x_s, x_t), dim=0)
y, f = model(x)
y_s, y_t = y.chunk(2, dim=0)
f_s, f_t = f.chunk(2, dim=0)
cls_loss = F.cross_entropy(y_s, labels_s)
transfer_loss = jmmd_loss(
(f_s, F.softmax(y_s, dim=1)),
(f_t, F.softmax(y_t, dim=1))
)
loss = cls_loss + transfer_loss * args.trade_off
cls_acc = accuracy(y_s, labels_s)[0]
tgt_acc = accuracy(y_t, labels_t)[0]
losses.update(loss.item(), x_s.size(0))
cls_accs.update(cls_acc.item(), x_s.size(0))
tgt_accs.update(tgt_acc.item(), x_t.size(0))
trans_losses.update(transfer_loss.item(), x_s.size(0))
# compute gradient and do SGD step
optimizer.zero_grad()
loss.backward()
optimizer.step()
lr_scheduler.step()
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
if i % args.print_freq == 0:
progress.display(i)
def main(args: argparse.Namespace):
logger = CompleteLogger(args.log, args.phase)
print(args)
if args.seed is not None:
random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.deterministic = True
warnings.warn('You have chosen to seed training. '
'This will turn on the CUDNN deterministic setting, '
'which can slow down your training considerably! '
'You may see unexpected behavior when restarting '
'from checkpoints.')
cudnn.benchmark = True
# Data loading code
train_transform = utils.get_train_transform(args.train_resizing, random_horizontal_flip=not args.no_hflip,
random_color_jitter=False, resize_size=args.resize_size,
norm_mean=args.norm_mean, norm_std=args.norm_std)
val_transform = utils.get_val_transform(args.val_resizing, resize_size=args.resize_size,
norm_mean=args.norm_mean, norm_std=args.norm_std)
print("train_transform: ", train_transform)
print("val_transform: ", val_transform)
train_source_dataset, train_target_dataset, val_dataset, test_dataset, num_classes, args.class_names = \
utils.get_dataset(args.data, args.root, args.source, args.target, train_transform, val_transform)
train_source_loader = DataLoader(train_source_dataset, batch_size=args.batch_size,
shuffle=True, num_workers=args.workers, drop_last=True)
train_target_loader = DataLoader(train_target_dataset, batch_size=args.batch_size,
shuffle=True, num_workers=args.workers, drop_last=True)
val_loader = DataLoader(val_dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.workers)
test_loader = DataLoader(test_dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.workers)
train_source_iter = ForeverDataIterator(train_source_loader)
train_target_iter = ForeverDataIterator(train_target_loader)
# create model
print("=> using model '{}'".format(args.arch))
backbone = utils.get_model(args.arch, pretrain=not args.scratch)
pool_layer = nn.Identity() if args.no_pool else None
classifier = ImageClassifier(backbone, num_classes, bottleneck_dim=args.bottleneck_dim,
pool_layer=pool_layer, finetune=not args.scratch).to(device)
# define loss function
if args.adversarial:
thetas = [Theta(dim).to(device) for dim in (classifier.features_dim, num_classes)]
else:
thetas = None
jmmd_loss = JointMultipleKernelMaximumMeanDiscrepancy(
kernels=(
[GaussianKernel(alpha=2 ** k) for k in range(-3, 2)],
(GaussianKernel(sigma=0.92, track_running_stats=False),)
),
linear=args.linear, thetas=thetas
).to(device)
parameters = classifier.get_parameters()
if thetas is not None:
parameters += [{"params": theta.parameters(), 'lr': 0.1} for theta in thetas]
# define optimizer
optimizer = SGD(parameters, args.lr, momentum=args.momentum, weight_decay=args.wd, nesterov=True)
lr_scheduler = LambdaLR(optimizer, lambda x: args.lr * (1. + args.lr_gamma * float(x)) ** (-args.lr_decay))
# resume from the best checkpoint
if args.phase != 'train':
checkpoint = torch.load(logger.get_checkpoint_path('best'), map_location='cpu')
classifier.load_state_dict(checkpoint)
# analysis the model
if args.phase == 'analysis':
# extract features from both domains
feature_extractor = nn.Sequential(classifier.backbone, classifier.pool_layer, classifier.bottleneck).to(device)
source_feature = collect_feature(train_source_loader, feature_extractor, device)
target_feature = collect_feature(train_target_loader, feature_extractor, device)
# plot t-SNE
tSNE_filename = osp.join(logger.visualize_directory, 'TSNE.pdf')
tsne.visualize(source_feature, target_feature, tSNE_filename)
print("Saving t-SNE to", tSNE_filename)
# calculate A-distance, which is a measure for distribution discrepancy
A_distance = a_distance.calculate(source_feature, target_feature, device)
print("A-distance =", A_distance)
return
if args.phase == 'test':
acc1 = utils.validate(test_loader, classifier, args, device)
print(acc1)
return
# start training
best_acc1 = 0.
for epoch in range(args.epochs):
# train for one epoch
train(train_source_iter, train_target_iter, classifier, jmmd_loss, optimizer,
lr_scheduler, epoch, args)
# evaluate on validation set
acc1 = utils.validate(val_loader, classifier, args, device)
# remember best acc@1 and save checkpoint
torch.save(classifier.state_dict(), logger.get_checkpoint_path('latest'))
if acc1 > best_acc1:
shutil.copy(logger.get_checkpoint_path('latest'), logger.get_checkpoint_path('best'))
best_acc1 = max(acc1, best_acc1)
print("best_acc1 = {:3.1f}".format(best_acc1))
# evaluate on test set
classifier.load_state_dict(torch.load(logger.get_checkpoint_path('best')))
acc1 = utils.validate(test_loader, classifier, args, device)
print("test_acc1 = {:3.1f}".format(acc1))
logger.close()
def main(args: argparse.Namespace):
if args.seed is not None:
random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.deterministic = True
warnings.warn('You have chosen to seed training. '
'This will turn on the CUDNN deterministic setting, '
'which can slow down your training considerably! '
'You may see unexpected behavior when restarting '
'from checkpoints.')
cudnn.benchmark = True
# Data loading code
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
train_transform = transforms.Compose([
ResizeImage(256),
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(), normalize
])
val_transform = transforms.Compose([
ResizeImage(256),
transforms.CenterCrop(224),
transforms.ToTensor(), normalize
])
dataset = datasets.__dict__[args.data]
train_source_dataset = dataset(root=args.root,
task=args.source,
download=True,
transform=train_transform)
train_source_loader = DataLoader(train_source_dataset,
batch_size=args.batch_size,
shuffle=True,
num_workers=args.workers,
drop_last=True)
train_target_dataset = dataset(root=args.root,
task=args.target,
download=True,
transform=train_transform)
train_target_loader = DataLoader(train_target_dataset,
batch_size=args.batch_size,
shuffle=True,
num_workers=args.workers,
drop_last=True)
val_dataset = dataset(root=args.root,
task=args.target,
download=True,
transform=val_transform)
val_loader = DataLoader(val_dataset,
batch_size=args.batch_size,
shuffle=False,
num_workers=args.workers)
if args.data == 'DomainNet':
test_dataset = dataset(root=args.root,
task=args.target,
evaluate=True,
download=True,
transform=val_transform)
test_loader = DataLoader(test_dataset,
batch_size=args.batch_size,
shuffle=False,
num_workers=args.workers)
else:
test_loader = val_loader
train_source_iter = ForeverDataIterator(train_source_loader)
train_target_iter = ForeverDataIterator(train_target_loader)
# create model
print("=> using pre-trained model '{}'".format(args.arch))
backbone = models.__dict__[args.arch](pretrained=True)
num_classes = train_source_dataset.num_classes
classifier = ImageClassifier(backbone, num_classes).to(device)
# define loss function
if args.adversarial:
thetas = [
Theta(dim).to(device)
for dim in (classifier.features_dim, num_classes)
]
else:
thetas = None
jmmd_loss = JointMultipleKernelMaximumMeanDiscrepancy(
kernels=([GaussianKernel(alpha=2**k) for k in range(-3, 2)],
(GaussianKernel(sigma=0.92, track_running_stats=False), )),
linear=args.linear,
thetas=thetas).to(device)
parameters = classifier.get_parameters()
if thetas is not None:
parameters += [{
"params": theta.parameters(),
'lr_mult': 0.1
} for theta in thetas]
# define optimizer
optimizer = SGD(parameters,
args.lr,
momentum=args.momentum,
weight_decay=args.wd,
nesterov=True)
lr_sheduler = StepwiseLR(optimizer,
init_lr=args.lr,
gamma=0.0003,
decay_rate=0.75)
# start training
best_acc1 = 0.
for epoch in range(args.epochs):
# train for one epoch
train(train_source_iter, train_target_iter, classifier, jmmd_loss,
optimizer, lr_sheduler, epoch, args)
# evaluate on validation set
acc1 = validate(val_loader, classifier, args)
# remember best acc@1 and save checkpoint
if acc1 > best_acc1:
best_model = copy.deepcopy(classifier.state_dict())
best_acc1 = max(acc1, best_acc1)
print("best_acc1 = {:3.1f}".format(best_acc1))
# evaluate on test set
classifier.load_state_dict(best_model)
acc1 = validate(test_loader, classifier, args)
print("test_acc1 = {:3.1f}".format(acc1))