def main():
args = parse_args()
global logger
logger = Logger(args.logpath)
torch.manual_seed(args.seed)
use_cuda = not args.no_cuda and torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
model = UNet2D(n_channels=1, n_classes=1).to(device)
optimizer = optim.SGD(model.parameters(),
lr=args.lr,
momentum=args.momentum)
criterion = SoftDiceLoss()
dataset = IRCAD2D(args.datapath, tissue='bone', binarymask=True)
print(f'Segmenting {dataset.tissue}')
if args.resume:
model, optimizer = load_checkpoint(model, optimizer, args.savepath,
args.savefile)
trainset, testset = train_valid_split(dataset)
trainloader = DataLoader(trainset, batch_size=args.batch_size)
testloader = DataLoader(testset, batch_size=args.batch_size)
train_meters = {
'loss': AverageMeter('trainloss', args.meterpath),
'dice': AverageMeter('traindice', args.meterpath)
}
test_meters = {
'loss': AverageMeter('testloss', args.meterpath),
'dice': AverageMeter('testdice', args.meterpath)
}
for epoch in range(1, args.epochs + 1):
train(args, model, device, trainloader, optimizer, epoch, train_meters,
criterion, savepath, savefile)
test(args, model, device, testloader, test_meters, epoch, criterion)
train_meters['loss'].save()
train_meters['dice'].save()
test_meters['loss'].save()
test_meters['dice'].save()
def main():
args = parse_args()
global logger
logger = Logger(args.logpath)
torch.manual_seed(args.seed)
start_gpu = f'cuda:{args.start_gpu}'
end_gpu = f'cuda:{args.end_gpu}'
model = ModelParallelUNet3D(n_channels=1, n_classes=1)
optimizer = optim.SGD(model.parameters(),
lr=args.lr,
momentum=args.momentum)
criterion = SoftDiceLoss()
dataset = IRCAD3D(args.datapath, tissue='bone')
print(f'Segmenting {dataset.tissue}')
trainset, testset = train_valid_split(dataset)
trainloader = DataLoader(trainset, batch_size=args.batch_size)
testloader = DataLoader(testset, batch_size=args.batch_size)
train_meters = {
'loss': AverageMeter('trainloss', args.meterpath),
'dice': AverageMeter('traindice', args.meterpath)
}
test_meters = {
'loss': AverageMeter('testloss', args.meterpath),
'dice': AverageMeter('testdice', args.meterpath)
}
for epoch in range(1, args.epochs + 1):
train(args, model, start_gpu, end_gpu, trainloader, optimizer, epoch,
train_meters, criterion)
test(args, model, start_gpu, end_gpu, testloader, test_meters, epoch,
criterion)
train_meters['loss'].save()
train_meters['dice'].save()
test_meters['loss'].save()
test_meters['dice'].save()
def main():
print(f'Kicking off.')
args = parse_args()
global logger
logger = Logger(args.logpath)
torch.manual_seed(args.seed)
use_cuda = not args.no_cuda and torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
model = UNet2D(n_channels=1, n_classes=1).to(device)
optimizer = optim.SGD(model.parameters(),
lr=args.lr,
momentum=args.momentum)
criterion = SoftDiceLoss()
print('loading data...')
dataset = AAPM(args.datapath)
trainset, testset = train_valid_split(dataset)
trainloader = DataLoader(trainset, batch_size=args.batch_size)
testloader = DataLoader(testset, batch_size=args.batch_size)
train_meters = {
'loss': AverageMeter('trainloss', args.meterpath),
'dice': AverageMeter('traindice', args.meterpath)
}
test_meters = {
'loss': AverageMeter('testloss', args.meterpath),
'dice': AverageMeter('testdice', args.meterpath)
}
for epoch in range(1, args.epochs + 1):
train(args, model, device, trainloader, optimizer, epoch, train_meters,
criterion)
test(args, model, device, testloader, test_meters, epoch, criterion)
train_meters['loss'].save()
train_meters['dice'].save()
test_meters['loss'].save()
test_meters['dice'].save()