def validate(epoch, val_loader, criterion, model, args):
""""""
batch_time = AverageMeter(name='batch_time')
loss_meter = AverageMeter(name='losses')
mapmeter = mAPMeter()
num_samples = len(val_loader)
model.eval()
end = time.time()
for batch_idx, (data, target) in enumerate(val_loader):
bs, n_crops, c, h, w = data.size()
data = data.view(-1, c, h, w)
if args.cuda:
data = data.cuda(non_blocking=True)
target = target.cuda(non_blocking=True)
output = model(data)
output = output.view(bs, n_crops, -1).mean(1)
loss = criterion(output, target)
loss_meter.update(loss.item(), data.size(0))
mapmeter.update(output, target)
if batch_idx % args.log_interval == 0 and batch_idx > 0:
log_progress('Validation', epoch, args.num_epochs, batch_idx, num_samples, batch_time, loss_meter, mapmeter)
if epoch == args.num_epochs:
del data, target
torch.cuda.empty_cache()
return loss_meter.avg
def main():
torch.backends.cudnn.benchmark = True
args = parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
# Data loading
loaders = XRayLoaders(data_dir=args.data, batch_size=args.batch_size)
val_loader = loaders.val_loader(imagetxt=args.valtxt)
encoder = DenseNet121()
model = TransferNet(encoder)
if args.resume:
if os.path.isfile(args.savefile):
print("=> loading checkpoint '{}'".format(args.savefile))
checkpoint = torch.load(args.savefile)
# must make sure the save states align.
save_state = get_save_state(checkpoint)
load_save_state(model, save_state)
print("=> loaded checkpoint '{}'".format(args.savefile))
else:
print("=> no checkpoint found at '{}'".format(args.savefile))
if args.parallel:
model = nn.DataParallel(model)
model = model.cuda()
if args.cuda and not args.parallel:
model.cuda()
optimizer = optim.Adadelta(model.parameters(), lr=args.lr)
criterion = nn.BCEWithLogitsLoss(size_average=True)
if args.cuda:
criterion.cuda()
val_meters = {
'val_loss': AverageMeter(name='valloss'),
'val_time': AverageMeter(name='valtime'),
'val_mavep': mAPMeter(),
'val_accuracy': AverageMeter(name='valaccuracy')
}
val_loss, val_map = validate(val_loader, criterion, model, val_meters,
args)
def main():
torch.backends.cudnn.benchmark = True
args = parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
# Data loading
loaders = XRayLoaders(data_dir=args.data, batch_size=args.batch_size)
train_loader = loaders.train_loader(imagetxt=args.traintxt)
val_loader = loaders.val_loader(imagetxt=args.valtxt)
model = LungXnet()
if args.parallel:
# model = nn.DataParallel(model)
model = model.cuda().half()
if args.cuda and not args.parallel:
model.cuda().half()
optimizer = optim.Adadelta(model.parameters(), lr=args.lr)
criterion = nn.BCEWithLogitsLoss(size_average=True)
if args.cuda:
criterion.cuda()
train_meters = {
'train_loss': AverageMeter(name='trainloss'),
'train_time': AverageMeter(name='traintime'),
'train_mavep': mAPMeter()
}
val_meters = {
'val_loss': AverageMeter(name='valloss'),
'val_time': AverageMeter(name='valtime'),
'val_mavep': mAPMeter()
}
epoch_time = AverageMeter(name='epoch_time')
end = time.time()
print(f'Number of epochs: {args.num_epochs}')
for epoch in range(1, args.num_epochs + 1):
print(f'epoch: {epoch}')
train_loss, train_map = train(train_loader,
optimizer,
criterion,
model,
train_meters,
args,
epoch=epoch)
val_loss, val_map = validate(val_loader,
criterion,
model,
val_meters,
args,
epoch=epoch)
epoch_time.update(time.time() - end)
end = time.time()
print(
f"\nJob's done! Total runtime: {epoch_time.sum}, Average runtime: {epoch_time.avg}"
)
def main():
torch.backends.cudnn.benchmark = True
args = parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
# Data loading
loaders = XRayLoaders(data_dir=args.data, batch_size=args.batch_size)
train_loader = loaders.train_loader(imagetxt=args.traintxt)
val_loader = loaders.val_loader(imagetxt=args.valtxt)
encoder = DenseNet121()
if args.resume:
if os.path.isfile(args.savefile):
print("=> loading checkpoint '{}'".format(args.savefile))
checkpoint = torch.load(args.savefile)
encoder.load_state_dict(checkpoint['state_dict'])
print("=> loaded checkpoint '{}'".format(args.savefile))
else:
print("=> no checkpoint found at '{}'".format(args.savefile))
model = TransferNet(encoder)
if args.parallel:
model = nn.DataParallel(model)
model = model.cuda()
if args.cuda and not args.parallel:
model.cuda()
optimizer = optim.Adadelta(model.parameters(), lr=args.lr)
criterion = nn.BCEWithLogitsLoss(size_average=True)
if args.cuda:
criterion.cuda()
train_meters = {
'train_loss': AverageMeter(name='trainloss'),
'train_time': AverageMeter(name='traintime'),
'train_mavep': mAPMeter()
}
val_meters = {
'val_loss': AverageMeter(name='valloss'),
'val_time': AverageMeter(name='valtime'),
'val_mavep': mAPMeter(),
'val_accuracy': AverageMeter(name='valaccuracy')
}
epoch_time = AverageMeter(name='epoch_time')
end = time.time()
print(f'Number of epochs: {args.num_epochs}')
for epoch in range(1, args.num_epochs + 1):
print(f'epoch: {epoch}')
train_loss, train_map = train(train_loader,
optimizer,
criterion,
model,
train_meters,
args,
epoch=epoch)
if epoch % 10 == 0:
save_checkpoint({
'epoch': epoch,
'state_dict': model.state_dict(),
'optimizer': optimizer.state_dict(),
})
val_loss, val_map = validate(val_loader,
criterion,
model,
val_meters,
args,
epoch=epoch)
epoch_time.update(time.time() - end)
end = time.time()
print(
f"\nJob's done! Total runtime: {epoch_time.sum}, Average runtime: {epoch_time.avg}"
)
train_meters['train_loss'].save('/home/ygx/lungs/lungs')
val_meters['val_loss'].save('/home/ygx/lungs/lungs')
val_meters['val_accuracy'].save('/home/ygx/lungs/lungs/acculogs')