dataset = Dataset_train(data_dir=data_dir,
fold=options.fold,
input_size=input_size,
normalize_mean=IMG_MEAN,
normalize_std=IMG_STD,
prob=options.prob)
trainloader = data.DataLoader(dataset,
batch_size=batch_size,
shuffle=True,
num_workers=options.workers)
# valset
# this only a quick val dataset where all images are 321*321.
valset = Dataset_val(data_dir=data_dir,
fold=options.fold,
input_size=input_size,
normalize_mean=IMG_MEAN,
normalize_std=IMG_STD)
valloader = data.DataLoader(valset,
batch_size=options.bs_val,
shuffle=False,
num_workers=options.workers,
drop_last=False)
save_pred_every = len(trainloader)
optimizer = optim.SGD([{
'params': get_10x_lr_params(model),
'lr': 10 * learning_rate
}],
lr=learning_rate,
optimizer = optim.SGD([{
'params': get_10x_lr_params(model),
'lr': 10 * learning_rate
}],
lr=learning_rate,
momentum=momentum,
weight_decay=weight_decay)
checkpoint_dir = 'checkpoint/fo=%d/' % options.fold
check_dir(checkpoint_dir)
# this only a quick val dataset where all images are 321*321.
valset = Dataset_val(data_dir=data_dir,
fold=options.fold,
input_size=input_size,
normalize_mean=IMG_MEAN,
normalize_std=IMG_STD)
valloader = data.DataLoader(valset,
batch_size=options.bs_val,
shuffle=False,
num_workers=4,
drop_last=False)
model.cuda()
model.eval()
valset.history_mask_list = [None] * 451
all_inter = np.zeros([5])
all_union = np.zeros([5])