def lmmd(source,
target,
s_label,
t_label,
kernel_mul=2.0,
kernel_num=5,
fix_sigma=None,
num_classes=10):
batch_size = source.size()[0]
weight_ss, weight_tt, weight_st = Weight.cal_weight(s_label,
t_label,
type='visual',
class_nums=num_classes)
weight_ss = torch.from_numpy(weight_ss).cuda()
weight_tt = torch.from_numpy(weight_tt).cuda()
weight_st = torch.from_numpy(weight_st).cuda()
kernels = guassian_kernel(source,
target,
kernel_mul=kernel_mul,
kernel_num=kernel_num,
fix_sigma=fix_sigma)
loss = torch.Tensor([0]).cuda()
#print(kernels)
if torch.sum(torch.isnan(sum(kernels))):
return loss
SS = kernels[:batch_size, :batch_size]
TT = kernels[batch_size:, batch_size:]
ST = kernels[:batch_size, batch_size:]
loss += torch.sum(weight_ss * SS + weight_tt * TT - 2 * weight_st * ST)
return loss
def conditional(self,
source,
target,
s_label,
t_label,
kernel_mul=2.0,
kernel_num=5,
fix_sigma=None):
batch_size = source.size()[0]
weight_ss, weight_tt, weight_st = Weight.cal_weight(s_label,
t_label,
type='visual')
weight_ss = torch.from_numpy(weight_ss).cuda()
weight_tt = torch.from_numpy(weight_tt).cuda()
weight_st = torch.from_numpy(weight_st).cuda()
kernels = self.guassian_kernel(source,
target,
kernel_mul=kernel_mul,
kernel_num=kernel_num,
fix_sigma=fix_sigma)
loss = torch.Tensor([0]).cuda()
if torch.sum(torch.isnan(sum(kernels))):
return loss
SS = kernels[:batch_size, :batch_size]
TT = kernels[batch_size:, batch_size:]
ST = kernels[:batch_size, batch_size:]
loss += torch.sum(weight_ss * SS + weight_tt * TT - 2 * weight_st * ST)
return loss
