def main(args):
global device
best_acc = 0.0
# create checkpoint dir and log dir
if not isdir(args.checkpoint):
print("Create new checkpoint folder " + args.checkpoint)
mkdir_p(args.checkpoint)
if not args.resume:
if isdir(args.logdir):
shutil.rmtree(args.logdir)
mkdir_p(args.logdir)
# create model
model = ROOTNET()
model.to(device)
optimizer = torch.optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
# optionally resume from a checkpoint
if args.resume:
if isfile(args.resume):
print("=> loading checkpoint '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
best_acc = checkpoint['best_acc']
model.load_state_dict(checkpoint['state_dict'])
optimizer.load_state_dict(checkpoint['optimizer'])
lr = optimizer.param_groups[0]['lr']
print("=> loaded checkpoint '{}' (epoch {})".format(
args.resume, checkpoint['epoch']))
else:
print("=> no checkpoint found at '{}'".format(args.resume))
cudnn.benchmark = True
print(' Total params: %.2fM' %
(sum(p.numel() for p in model.parameters()) / 1000000.0))
train_loader = DataLoader(GraphDataset(root=args.train_folder),
batch_size=args.train_batch,
shuffle=True,
follow_batch=['joints'])
val_loader = DataLoader(GraphDataset(root=args.val_folder),
batch_size=args.test_batch,
shuffle=False,
follow_batch=['joints'])
test_loader = DataLoader(GraphDataset(root=args.test_folder),
batch_size=args.test_batch,
shuffle=False,
follow_batch=['joints'])
if args.evaluate:
print('\nEvaluation only')
test_loss, test_acc = test(test_loader, model)
print('test_loss {:.8f}. test_acc: {:.6f}'.format(test_loss, test_acc))
return
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer,
args.schedule,
gamma=args.gamma)
logger = SummaryWriter(log_dir=args.logdir)
for epoch in range(args.start_epoch, args.epochs):
lr = scheduler.get_last_lr()
print('\nEpoch: %d | LR: %.8f' % (epoch + 1, lr[0]))
train_loss = train(train_loader, model, optimizer, args)
val_loss, val_acc = test(val_loader, model)
test_loss, test_acc = test(test_loader, model)
scheduler.step()
print('Epoch{:d}. train_loss: {:.6f}.'.format(epoch + 1, train_loss))
print('Epoch{:d}. val_loss: {:.6f}. val_acc: {:.6f}'.format(
epoch + 1, val_loss, val_acc))
print('Epoch{:d}. test_loss: {:.6f}. test_acc: {:.6f}'.format(
epoch + 1, test_loss, test_acc))
# remember best acc and save checkpoint
is_best = val_acc > best_acc
best_acc = max(val_acc, best_acc)
save_checkpoint(
{
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'best_acc': best_acc,
'optimizer': optimizer.state_dict()
},
is_best,
checkpoint=args.checkpoint)
info = {
'train_loss': train_loss,
'val_loss': val_loss,
'val_accuracy': val_acc,
'test_loss': test_loss,
'test_accuracy': test_acc
}
for tag, value in info.items():
logger.add_scalar(tag, value, epoch + 1)
print("=> loading checkpoint '{}'".format(
os.path.join(args.checkpoint, 'model_best.pth.tar')))
checkpoint = torch.load(os.path.join(args.checkpoint,
'model_best.pth.tar'))
best_epoch = checkpoint['epoch']
model.load_state_dict(checkpoint['state_dict'])
print("=> loaded checkpoint '{}' (epoch {})".format(
os.path.join(args.checkpoint, 'model_best.pth.tar'), best_epoch))
test_loss, test_acc = test(test_loader, model)
print('Best epoch:\n test_loss {:8f} test_acc {:8f}'.format(
test_loss, test_acc))
def main(args):
global device
lowest_loss = 1e20
# create checkpoint dir and log dir
if not isdir(args.checkpoint):
print("Create new checkpoint folder " + args.checkpoint)
mkdir_p(args.checkpoint)
if not args.resume:
if isdir(args.logdir):
shutil.rmtree(args.logdir)
mkdir_p(args.logdir)
# create model
model = PairCls()
model.to(device)
optimizer = torch.optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
# optionally resume from a checkpoint
if args.resume:
if isfile(args.resume):
print("=> loading checkpoint '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
model.load_state_dict(checkpoint['state_dict'])
optimizer.load_state_dict(checkpoint['optimizer'])
args.start_epoch = checkpoint['epoch']
lowest_loss = checkpoint['lowest_loss']
print("=> loaded checkpoint '{}' (epoch {})".format(
args.resume, checkpoint['epoch']))
else:
print("=> no checkpoint found at '{}'".format(args.resume))
cudnn.benchmark = True
print(' Total params: %.2fM' %
(sum(p.numel() for p in model.parameters()) / 1000000.0))
train_loader = DataLoader(GraphDataset(root=args.train_folder),
batch_size=args.train_batch,
shuffle=True)
val_loader = DataLoader(GraphDataset(root=args.val_folder),
batch_size=args.test_batch,
shuffle=False)
test_loader = DataLoader(GraphDataset(root=args.test_folder),
batch_size=args.test_batch,
shuffle=False)
if args.evaluate:
print('\nEvaluation only')
test_loss = test(test_loader,
model,
args,
save_result=True,
best_epoch=args.start_epoch)
print('test_loss {:8f}'.format(test_loss))
return
lr = args.lr
logger = SummaryWriter(log_dir=args.logdir)
for epoch in range(args.start_epoch, args.epochs):
print('\nEpoch: %d | LR: %.8f' % (epoch + 1, lr))
lr = adjust_learning_rate(optimizer, epoch, lr, args.schedule,
args.gamma)
train_loss = train(train_loader, model, optimizer, args)
val_loss = test(val_loader, model, args)
test_loss = test(test_loader, model, args, best_epoch=epoch + 1)
print('Epoch{:d}. train_loss: {:.6f}.'.format(epoch, train_loss))
print('Epoch{:d}. val_loss: {:.6f}.'.format(epoch, val_loss))
print('Epoch{:d}. test_loss: {:.6f}.'.format(epoch, test_loss))
# remember best acc and save checkpoint
is_best = val_loss < lowest_loss
lowest_loss = min(val_loss, lowest_loss)
save_checkpoint(
{
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'lowest_loss': lowest_loss,
'optimizer': optimizer.state_dict()
},
is_best,
checkpoint=args.checkpoint)
info = {
'train_loss': train_loss,
'val_loss': val_loss,
'test_loss': test_loss
}
for tag, value in info.items():
logger.add_scalar(tag, value, epoch + 1)
print("=> loading checkpoint '{}'".format(
os.path.join(args.checkpoint, 'model_best.pth.tar')))
checkpoint = torch.load(os.path.join(args.checkpoint,
'model_best.pth.tar'))
best_epoch = checkpoint['epoch']
model.load_state_dict(checkpoint['state_dict'])
print("=> loaded checkpoint '{}' (epoch {})".format(
os.path.join(args.checkpoint, 'model_best.pth.tar'), best_epoch))
test_loss = test(test_loader,
model,
args,
save_result=True,
best_epoch=best_epoch)
print('Best epoch:\n test_loss {:8f}'.format(test_loss))
def main(args):
global device
lowest_loss = 1e20
# create checkpoint dir and log dir
if not isdir(args.checkpoint):
print("Create new checkpoint folder " + args.checkpoint)
mkdir_p(args.checkpoint)
if not args.resume:
if isdir(args.logdir):
shutil.rmtree(args.logdir)
mkdir_p(args.logdir)
# create model
model = models.__dict__["skinnet"](args.nearest_bone, args.Dg, args.Lf)
model.to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr, weight_decay=args.weight_decay)
# optionally resume from a checkpoint
lr = args.lr
if args.resume:
if isfile(args.resume):
print("=> loading checkpoint '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
lowest_loss = checkpoint['lowest_loss']
model.load_state_dict(checkpoint['state_dict'])
optimizer.load_state_dict(checkpoint['optimizer'])
lr = optimizer.param_groups[0]['lr']
print("=> loaded checkpoint '{}' (epoch {})".format(args.resume, checkpoint['epoch']))
else:
print("=> no checkpoint found at '{}'".format(args.resume))
cudnn.benchmark = True
print(' Total params: %.2fM' % (sum(p.numel() for p in model.parameters()) / 1000000.0))
train_loader = DataLoader(SkinDataset(root=args.train_folder), batch_size=args.train_batch, shuffle=True)
val_loader = DataLoader(SkinDataset(root=args.val_folder), batch_size=args.test_batch, shuffle=False)
test_loader = DataLoader(SkinDataset(root=args.test_folder), batch_size=args.test_batch, shuffle=False)
if args.evaluate:
print('\nEvaluation only')
test_loss = test(test_loader, model, args, save_result=True)
print('test_loss {:6f}'.format(test_loss))
return
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, args.schedule, gamma=args.gamma)
logger = SummaryWriter(log_dir=args.logdir)
for epoch in range(args.start_epoch, args.epochs):
lr = scheduler.get_last_lr()
print('\nEpoch: %d | LR: %.8f' % (epoch + 1, lr[0]))
train_loss = train(train_loader, model, optimizer, args)
val_loss = test(val_loader, model, args)
test_loss = test(test_loader, model, args)
scheduler.step()
print('Epoch{:d}. train_loss: {:.6f}.'.format(epoch + 1, train_loss))
print('Epoch{:d}. val_loss: {:.6f}.'.format(epoch + 1, val_loss))
print('Epoch{:d}. test_loss: {:.6f}.'.format(epoch + 1, test_loss))
# remember best acc and save checkpoint
is_best = val_loss < lowest_loss
lowest_loss = min(val_loss, lowest_loss)
save_checkpoint({'epoch': epoch + 1, 'state_dict': model.state_dict(), 'lowest_loss': lowest_loss,
'optimizer': optimizer.state_dict()}, is_best, checkpoint=args.checkpoint)
info = {'train_loss': train_loss, 'val_loss': val_loss, 'test_loss': test_loss}
for tag, value in info.items():
logger.add_scalar(tag, value, epoch + 1)