计算损失时出错 - 不支持的操作数类型 -：'tuple' 和 'tuple'

如何解决计算损失时出错 - 不支持的操作数类型 -：'tuple' 和 'tuple'

我正在尝试使用以下代码计算损失

class MSE_gradient(nn.Module):
  def __init__(self):
    super(MSE_gradient,self).__init__()

  def forward(self,inputs,targets):
    diff = torch.gradient(inputs) - torch.gradient(targets)
    return torch.sum(diff*diff)/diff.numel()

但我收到错误

      6   def forward(self,targets):
----> 7     diff = torch.gradient(inputs) - torch.gradient(targets)
      8     return torch.sum(diff*diff)/diff.numel()

TypeError: unsupported operand type(s) for -: 'tuple' and 'tuple'

我正在尝试计算图像的梯度并找到它的 MSE 误差。请帮忙

解决方法

torch.gradient 输出一个元组，如您所见，您不能从另一个元素中减去一个元组，因此出现错误。

用 diff = torch.gradient(inputs) - torch.gradient(targets) 代替

diff = [] 
for a,b in zip (torch.gradient(inputs),torch.gradient(targets)):
    diff.append(a-b)

# if you need tuple,cast diff as tuple
diff = tuple(diff)