def forward(self, source, target):
loss = 0
source = self.sharedNet(source)
#print ("Pass shared net")
if (self.training == True):
target = self.sharedNet(target)
loss += mmd.mmd_rbf_noaccelerate(source, target)
source = self.cls_fc(source)
return source, loss
def forward(self, source, target):
loss = 0
source = self.sharedNet(source)
if self.training == True:
target = self.sharedNet(target)
#loss += mmd.mmd_rbf_accelerate(source, target)
loss += mmd.mmd_rbf_noaccelerate(source, target)
source = self.cls_fc(source)
#target = self.cls_fc(target)
return source, loss
def forward(self, source, target, s_label, mu):
source = self.sharedNet(source)
target = self.sharedNet(target)
source_pred, cmmd_loss = self.Inception(source, target, s_label)
mmd_loss = 0
if self.training:
source = self.avgpool(source)
source = source.view(source.size(0), -1)
target = self.avgpool(target)
target = target.view(target.size(0), -1)
mmd_loss += mmd.mmd_rbf_noaccelerate(source, target)
loss = (1 - mu) * cmmd_loss + mu * mmd_loss
return source_pred, loss
def loss_adapt_fun(X, Y, loss_tran_type):
if loss_tran_type == 'mk_mmd':
#mmd_loss = mmd.MMD_loss(kernel_type='rbf', kernel_mul=1, kernel_num=10)
#loss = mmd_loss(X, Y)
loss = mmd.mmd_rbf_noaccelerate(X,
Y,
kernel_mul=2.0,
kernel_num=5,
fix_sigma=None)
elif loss_tran_type == 'mmd':
#mmd_loss = mmd.MMD_loss(kernel_type='rbf', kernel_mul=1, kernel_num=1)
#loss = mmd_loss(X, Y)
loss = mmd.mmd_rbf_noaccelerate(X,
Y,
kernel_mul=2.0,
kernel_num=1,
fix_sigma=None)
#loss = mmd.mmd_rbf_loss(X, Y)
else:
loss = 0
return loss
def forward(self, source, target):
loss = 0
# 经过ResNet预训练
source = self.sharedNet(source)
if self.training == True:
target = self.sharedNet(target)
#loss += mmd.mmd_rbf_accelerate(source, target)
# 损失函数 源域目标域的mmd距离
loss += mmd.mmd_rbf_noaccelerate(source, target)
# 对源域做了一个线性变换然后返回回去 做src pred（经过网络后的预测值）
source = self.cls_fc(source)
#target = self.cls_fc(target)
return source, loss
def forward(self, source, target):
loss = 0
source = self.sharedNet(source)
if self.training == True:
target = self.sharedNet(target)
# mmd_loss = mmd.mmd_rbf_accelerate(source, target)
mmd_loss = mmd.mmd_rbf_noaccelerate(source, target)
loss += mmd_loss
if self.print_mmd_loss:
print('mmd loss:', mmd_loss.data[0])
pass
source = self.cls_fc(source)
#target = self.cls_fc(target)
return source, loss
def forward(self, src, tar):
loss = 0
x_src = self.featureCap(src)
x_src_mmd = x_src.view(x_src.size(0), -1)
# 这里为了设置全连接层的输入神经元个数，故需要展示平坦层的特征数（=全连接层输入神经元个数）
# print(x_src_mmd.size(1))
if self.training == True:
x_tar = self.featureCap(tar)
x_tar_mmd = x_tar.view(x_tar.size(0), -1)
#loss += mmd.mmd_rbf_accelerate(source, target)
loss += mmd.mmd_rbf_noaccelerate(x_src_mmd, x_tar_mmd)
y_src = self.fc1(x_src_mmd)
y_src = self.fc2(y_src)
#target = self.cls_fc(target)
return y_src, loss
def forward(self, source, target):
loss = 0
img_rec_s = 0
img_rec_t = 0
source = self.sharedNet(source)
if self.training == True:
target = self.sharedNet(target)
#loss += mmd.mmd_rbf_accelerate(source, target)
loss += mmd.mmd_rbf_noaccelerate(source, target)
feat_encode = self.rec_dense(source)
feat_encode = feat_encode.view(-1, 512, 7, 7)
img_rec_s = self.rec_feat(feat_encode)
feat_encode = self.rec_dense(target)
feat_encode = feat_encode.view(-1, 512, 7, 7)
img_rec_t = self.rec_feat(feat_encode)
source = self.cls_fc(source)
#target = self.cls_fc(target)
return source, loss, img_rec_s, img_rec_t