flag_simu=opt['loader'])
trainLoader_ICH = data.DataLoader(dataLoader_train_ICH,
batch_size=batch_size,
shuffle=True)
epoch = 0
gen_iterations = 1
display_iters = 5
loss1_sum, loss2_sum = 0, 0
Validation_loss = []
loss_L1 = lossL1()
while epoch < niter:
epoch += 1
# training phase
unet3d.train()
for idx, (rdf_inputs, rdfs, masks, weights, wGs,
D) in enumerate(trainLoader_ICH):
# training of unet3d
rdf_inputs0 = (rdf_inputs.to(device0, dtype=torch.float) +
trans) * scale
rdfs0 = (rdfs.to(device0, dtype=torch.float) + trans) * scale
masks0 = masks.to(device0, dtype=torch.float)
weights0 = weights.to(device0, dtype=torch.float)
wGs0 = wGs.to(device0, dtype=torch.float)
if epoch == 1:
QSM = unet3d(rdf_inputs0)[:, 0, ...]
QSM = np.squeeze(np.asarray(QSM.cpu().detach()))
adict = {}
adict['QSM'] = QSM
flag_RDF_input=1)
trainLoader_ICH = data.DataLoader(dataLoader_train_ICH,
batch_size=batch_size,
shuffle=True)
epoch = 0
gen_iterations = 1
display_iters = 5
loss1_sum, loss2_sum = 0, 0
Validation_loss = []
loss_L1 = lossL1()
while epoch < niter:
epoch += 1
# training phase
unet3d.train(), resnet.train()
for idx, (rdf_inputs, rdfs, masks, weights, wGs,
D) in enumerate(trainLoader_ICH):
# training of unet3d
rdf_inputs0 = rdf_inputs.to(device0, dtype=torch.float)
rdfs0 = rdfs.to(device0, dtype=torch.float)
masks0 = masks.to(device0, dtype=torch.float)
weights0 = weights.to(device0, dtype=torch.float)
wGs0 = wGs.to(device0, dtype=torch.float)
loss_fidelity = BayesianQSM_train(model=unet3d,
input_RDFs=rdf_inputs0,
in_loss_RDFs=rdfs0,
QSMs=0,
Masks=masks0,
fidelity_Ws=weights0,
use_deconv=1,
use_deconv2=1,
renorm=0,
flag_r_train=0,
flag_UTFI=1)
model.to(device)
# optimizer
optimizer = optim.Adam(model.parameters(), lr=lr, betas=(0.5, 0.999))
Validation_loss = []
loss_total_list = []
while epoch < niter:
# training phase
model.train()
for idx, (ifreqs, masks, data_weights, wGs) in enumerate(trainLoader):
ifreqs = ifreqs.to(device, dtype=torch.float)
masks = masks.to(device, dtype=torch.float)
data_weights = data_weights.to(device, dtype=torch.float)
wGs = wGs.to(device, dtype=torch.float)
loss_PDF, loss_fidelity, loss_tv = utfi_train(model,
optimizer,
ifreqs,
masks,
data_weights,
wGs,
D,
D_smv,
lambda_pdf,
