def get_current_visuals(self):
real_A = util.tensor2im(self.input_A)
fake_B = util.tensor2im(self.fake_B)
rec_A = util.tensor2im(self.rec_A)
real_B = util.tensor2im(self.input_B)
fake_A = util.tensor2im(self.fake_A)
rec_B = util.tensor2im(self.rec_B)
real_Ay=util.tensor2im_hd_neck(self.input_A_y)
fake_B_local=util.tensor2im(self.fake_B_local)
real_B_local=util.tensor2im(self.real_B)
#Attention map for Real MRI
att_real_B=util.tensor2im(self.att_map_real_B_struct)
#Segmentation probability map for pseudoMRI
att_fake_B=util.tensor2im(self.att_map_fake_B_struct)
if self.opt.Net=='Split-Unet':
pred_B=self.netSeg.forward_1(self.real_B)
else:
pred_B=self.netSeg(self.real_B)
pred_B=torch.argmax(pred_B, dim=1)
pred_B=pred_B.view(1,1,256,256)
pred_B=pred_B.data
seg_B=util.tensor2im_hd_neck(pred_B)#() #
ret_visuals = OrderedDict([('real_A', real_A), ('fake_B', fake_B), ('rec_A', rec_A),
('real_B', real_B), ('fake_A', fake_A), ('rec_B', rec_B),
('real_B_seg', seg_B),
('real_A_GT_seg',real_Ay),
])
return ret_visuals
def get_current_visuals(self):
real_A = util.tensor2im(self.input_A)
real_Ay = util.tensor2im_hd_neck(self.input_A_y)
pred_A = self.netSeg_A(self.input_A)
pred_A = torch.argmax(pred_A, dim=1)
pred_A = pred_A.view(self.input_A.size()[0], 1, 256, 256)
pred_A = pred_A.data
seg_A = util.tensor2im_hd_neck(pred_A) #
ret_visuals = OrderedDict([('real_A', real_A),
('real_A_GT_seg', real_Ay),
('real_A_seg', seg_A)])
return ret_visuals
def get_current_visuals(self):
real_A = util.tensor2im(self.input_A)
#fake_B = util.tensor2im(self.fake_B)
#fake_B[fake_B>450]=450
#rec_A = util.tensor2im(self.rec_A)
#real_B = util.tensor2im(self.input_B)
#fake_A = util.tensor2im(self.fake_A)
#rec_B = util.tensor2im(self.rec_B)
real_Ay = util.tensor2im_hd_neck(self.input_A_y)
#real_A_AB_CT=torch.cat((self.real_A,self.fake_B),1)
pred_A, _, rough = self.netSeg_A(self.input_A)
pred_A = F.softmax(pred_A, dim=1)
rough = F.softmax(rough, dim=1)
LP_refine = pred_A[0, 1, :, :]
LP_rough = rough[0, 1, :, :]
LP_rough = LP_rough.view(-1, 1, 256, 256)
LP_refine = LP_refine.view(-1, 1, 256, 256)
LP_rough = LP_rough.data
LP_refine = LP_refine.data
LP_rough, d999 = self.tensor2im_jj(LP_rough)
LP_refine, d999 = self.tensor2im_jj(LP_refine)
pred_A = torch.argmax(pred_A, dim=1)
pred_A = pred_A.view(self.input_A.size()[0], 1, 256, 256)
pred_A = pred_A.data
#global_map=global_map.data
seg_A = util.tensor2im_hd_neck(pred_A) #
#global_map=util.tensor2im(global_map)
#pred_B=self.netSeg_B(self.real_B)
#fake_B_seg=self.netSeg_B(self.fake_B)
#pred_B=pred_B.data
#fake_B_seg=fake_B_seg.data
#seg_B=util.tensor2im(pred_B) #
#fake_B_seg=util.tensor2im(fake_B_seg) #
ret_visuals = OrderedDict([('real_A', real_A),
('real_A_GT_seg', real_Ay),
('real_A_seg', seg_A),
('LP_rough', LP_rough),
('LP_refine', LP_refine)])
#if self.opt.isTrain and self.opt.identity > 0.0:
# ret_visuals['idt_A'] = util.tensor2im(self.idt_A)
# ret_visuals['idt_B'] = util.tensor2im(self.idt_B)
return ret_visuals
def net_G_A_A2B_Segtest_image_varian(self):
self.test_A = self.test_A.cuda()
self.test_AB = self.test_AB.cuda()
self.test_A_y = self.test_A_y.cuda()
#self.test_AB=self.netG_A(self.test_A)
test_img = self.test_A
#test_img=torch.cat((self.test_A,self.test_AB),1)
#print (self.test_A.size())
#print (self.test_AB.size())
#print (test_img.size())
test_img = test_img.float()
A_AB_seg = self.netSeg_A(test_img)
#A_AB_seg=F.softmax(A_AB_seg,dim=1)
loss = self.dice_loss(A_AB_seg, self.test_A_y)
#print (A_AB_seg.size())
A_AB_seg = F.softmax(A_AB_seg, dim=1)
LP_map = A_AB_seg[:, 1, :, :]
RP_map = A_AB_seg[:, 2, :, :]
LS_map = A_AB_seg[:, 3, :, :]
RS_map = A_AB_seg[:, 4, :, :]
LBP_map = A_AB_seg[:, 5, :, :]
RBP_map = A_AB_seg[:, 6, :, :]
Sub_map = A_AB_seg[:, 7, :, :]
Spine_map = A_AB_seg[:, 8, :, :]
RP_map = RP_map.view(1, 1, 256, 256)
LP_map = LP_map.view(1, 1, 256, 256)
LS_map = LS_map.view(1, 1, 256, 256)
RS_map = RS_map.view(1, 1, 256, 256)
Sub_map = Sub_map.view(1, 1, 256, 256)
LBP_map = LBP_map.view(1, 1, 256, 256)
RBP_map = RBP_map.view(1, 1, 256, 256)
Spine_map = Spine_map.view(1, 1, 256, 256)
RP_map = RP_map.data
LP_map = LP_map.data
LS_map = LS_map.data
RS_map = RS_map.data
Sub_map = Sub_map.data
LBP_map = LBP_map.data
RBP_map = RBP_map.data
Spine_map = Spine_map.data
RP_map_tp = RP_map
RP_map_tp = RP_map_tp + LP_map
RP_map_tp = RP_map_tp + LS_map
RP_map_tp = RP_map_tp + RS_map
RP_map_tp = RP_map_tp + Sub_map
RP_map_tp = RP_map_tp + LBP_map
RP_map_tp = RP_map_tp + RBP_map
RP_map_tp = RP_map_tp + Spine_map
d0, d1, d2, d3, d4, d5, d6, d7, d8 = self.cal_dice_loss_PDDA_val(
A_AB_seg, self.test_A_y)
A_AB_seg = torch.argmax(A_AB_seg, dim=1)
A_AB_seg = A_AB_seg.view(1, 1, 256, 256)
A_AB_seg_out = A_AB_seg.data
#print (A_AB_seg_out.size())
#A_AB_seg_out=torch.argmax(A_AB_seg_out, dim=1)
A_AB_seg_out = A_AB_seg_out.view(1, 1, 256, 256)
#print (A_AB_seg_out.size())
A_y_out = self.test_A_y.data
self.test_A_y = self.test_A_y.cuda()
test_A_data = self.test_A.data
#test_AB=self.test_AB.data
A_AB_seg = A_AB_seg.data
A_y = self.test_A_y.data
test_A_data, d999 = self.tensor2im_jj(test_A_data)
LP_map_data, d999 = self.tensor2im_jj(LP_map)
RP_map_data, d999 = self.tensor2im_jj(RP_map)
RP_map_tp_data, d999 = self.tensor2im_jj(RP_map_tp)
#test_AB,d2=self.tensor2im_jj(test_AB)
A_AB_seg = util.tensor2im_hd_neck(A_AB_seg)
A_y = util.tensor2im_hd_neck(A_y)
#print (test_A_data.shape)
#print (test_AB.shape)
#print (A_AB_seg.shape)
test_A_data = test_A_data[:, 256:512, :]
LP_map_data = LP_map_data[:, 256:512, :]
RP_map_data = RP_map_data[:, 256:512, :]
RP_map_tp_data = RP_map_tp_data[:, 256:512, :]
#test_AB=test_AB[:,256:512,:]
A_AB_seg = A_AB_seg #[:,256:512,:]
A_y = A_y #[:,256:512,:]
#test_A_data=test_A_data[:,512:1024,:]
#test_AB=test_AB[:,512:1024,:]
#A_AB_seg=A_AB_seg[:,512:1024,:]
#A_y=A_y[:,512:1024,:]
image_numpy_all = np.concatenate((
test_A_data,
A_y,
), axis=1)
#image_numpy_all=np.concatenate((image_numpy_all,A_y,),axis=1)
image_numpy_all = np.concatenate((
image_numpy_all,
A_AB_seg,
), axis=1)
#image_numpy_all=np.concatenate((image_numpy_all,LP_map_data,),axis=1)
#RP_map_data=RP_map_data+LP_map_data
image_numpy_all = np.concatenate((
image_numpy_all,
RP_map_tp_data,
),
axis=1)
#self.fake_A2B=self.fake_A2B.data
#self.fake_A2B_A_img, output_fakeA2B=self.tensor2im_jj(self.fake_A2B)
#print (loss)
return loss, self.test_A.cpu().float().numpy(), A_AB_seg_out.cpu(
).float().numpy(), A_y_out.cpu().float().numpy(
), image_numpy_all, d0, d1, d2, d3, d4, d5, d6, d7, d8, LP_map.cpu(
).float().numpy(), RP_map.cpu().float().numpy(), LS_map.cpu().float(
).numpy(), RS_map.cpu().float().numpy(), Sub_map.cpu().float().numpy(
), RP_map_tp.cpu().float().numpy()
def net_G_A_A2B_Segtest_image(self):
self.test_A = self.test_A.cuda()
self.test_AB = self.test_AB.cuda()
self.test_A_y = self.test_A_y.cuda()
#self.test_AB=self.netG_A(self.test_A)
test_img = self.test_A
#test_img=torch.cat((self.test_A,self.test_AB),1)
#print (self.test_A.size())
#print (self.test_AB.size())
#print (test_img.size())
test_img = test_img.float()
A_AB_seg, _, global_map = self.netSeg_A(test_img)
#A_AB_seg=F.softmax(A_AB_seg,dim=1)
loss = self.dice_loss(A_AB_seg, self.test_A_y)
#print (A_AB_seg.size())
A_AB_seg = F.softmax(A_AB_seg, dim=1)
LP_map = A_AB_seg[:, 1, :, :]
RP_map = A_AB_seg[:, 2, :, :]
RP_map = RP_map.view(1, 1, 256, 256)
LP_map = LP_map.view(1, 1, 256, 256)
RP_map = RP_map.data
LP_map = LP_map.data
global_map = global_map.data
A_AB_seg = torch.argmax(A_AB_seg, dim=1)
A_AB_seg = A_AB_seg.view(1, 1, 256, 256)
A_AB_seg_out = A_AB_seg.data
#print (A_AB_seg_out.size())
#A_AB_seg_out=torch.argmax(A_AB_seg_out, dim=1)
A_AB_seg_out = A_AB_seg_out.view(1, 1, 256, 256)
#print (A_AB_seg_out.size())
A_y_out = self.test_A_y.data
self.test_A_y = self.test_A_y.cuda()
test_A_data = self.test_A.data
#test_AB=self.test_AB.data
A_AB_seg = A_AB_seg.data
A_y = self.test_A_y.data
test_A_data, d999 = self.tensor2im_jj(test_A_data)
LP_map_data, d999 = self.tensor2im_jj(LP_map)
RP_map_data, d999 = self.tensor2im_jj(RP_map)
#global_map,d999=self.tensor2im_jj(global_map)
#test_AB,d2=self.tensor2im_jj(test_AB)
A_AB_seg = util.tensor2im_hd_neck(A_AB_seg)
A_y = util.tensor2im_hd_neck(A_y)
#print (test_A_data.shape)
#print (test_AB.shape)
#print (A_AB_seg.shape)
test_A_data = test_A_data[:, 256:512, :]
LP_map_data = LP_map_data[:, 256:512, :]
RP_map_data = RP_map_data[:, 256:512, :]
#global_map=global_map[:,256:512,:]
#test_AB=test_AB[:,256:512,:]
A_AB_seg = A_AB_seg #[:,256:512,:]
A_y = A_y #[:,256:512,:]
#test_A_data=test_A_data[:,512:1024,:]
#test_AB=test_AB[:,512:1024,:]
#A_AB_seg=A_AB_seg[:,512:1024,:]
#A_y=A_y[:,512:1024,:]
image_numpy_all = np.concatenate((
test_A_data,
A_y,
), axis=1)
#image_numpy_all=np.concatenate((image_numpy_all,A_y,),axis=1)
image_numpy_all = np.concatenate((
image_numpy_all,
A_AB_seg,
), axis=1)
#image_numpy_all=np.concatenate((image_numpy_all,global_map,),axis=1)
image_numpy_all = np.concatenate((
image_numpy_all,
LP_map_data,
),
axis=1)
image_numpy_all = np.concatenate((
image_numpy_all,
RP_map_data,
),
axis=1)
#self.fake_A2B=self.fake_A2B.data
#self.fake_A2B_A_img, output_fakeA2B=self.tensor2im_jj(self.fake_A2B)
#print (loss)
return loss, self.test_A.cpu().float().numpy(), A_AB_seg_out.cpu(
).float().numpy(), A_y_out.cpu().float().numpy(), image_numpy_all
def net_G_A_A2B_Segtest_image(self):
self.test_A = self.test_A.cuda()
self.test_AB = self.test_AB.cuda()
self.test_A_y = self.test_A_y.cuda()
#self.test_AB=self.netG_A(self.test_A)
test_img = self.test_A
#test_img=torch.cat((self.test_A,self.test_AB),1)
#print (self.test_A.size())
#print (self.test_AB.size())
#print (test_img.size())
test_img = test_img.float()
A_AB_seg = self.netSeg_A(test_img)
#A_AB_seg=F.softmax(A_AB_seg,dim=1)
#loss=self.dice_loss(A_AB_seg,self.test_A_y)
loss = 0
#print (A_AB_seg.size())
A_AB_seg = F.softmax(A_AB_seg, dim=1)
LP_map = A_AB_seg[:, 1, :, :]
RP_map = A_AB_seg[:, 2, :, :]
RP_map = RP_map.view(1, 1, 256, 256)
LP_map = LP_map.view(1, 1, 256, 256)
RP_map = RP_map.data
LP_map = LP_map.data
A_AB_seg = torch.argmax(A_AB_seg, dim=1)
A_AB_seg = A_AB_seg.view(1, 1, 256, 256)
A_AB_seg_out = A_AB_seg.data
#print (A_AB_seg_out.size())
#A_AB_seg_out=torch.argmax(A_AB_seg_out, dim=1)
A_AB_seg_out = A_AB_seg_out.view(1, 1, 256, 256)
#print (A_AB_seg_out.size())
A_y_out = self.test_A_y.data
self.test_A_y = self.test_A_y.cuda()
test_A_data = self.test_A.data
#test_AB=self.test_AB.data
A_AB_seg = A_AB_seg.data
A_y = self.test_A_y.data
test_A_data, d999 = self.tensor2im_jj(test_A_data)
#test_AB,d2=self.tensor2im_jj(test_AB)
A_AB_seg = util.tensor2im_hd_neck(A_AB_seg)
A_y = util.tensor2im_hd_neck(A_y)
test_A_data = test_A_data[:, 256:512, :]
#test_AB=test_AB[:,256:512,:]
A_AB_seg = A_AB_seg #[:,256:512,:]
A_y = A_y #[:,256:512,:]
image_numpy_all = np.concatenate((
test_A_data,
A_y,
), axis=1)
#image_numpy_all=np.concatenate((image_numpy_all,A_y,),axis=1)
image_numpy_all = np.concatenate((
image_numpy_all,
A_AB_seg,
), axis=1)
return loss, self.test_A.cpu().float().numpy(), A_AB_seg_out.cpu(
).float().numpy(), A_y_out.cpu().float().numpy(), image_numpy_all