python - Pytorch 中的孪生神经网络

Question

如何在 PyTorch 中实现孪生神经网络？

What is a siamese neural network? A siamese neural network consists in two identical neural networks, each one taking one input. Identical means that the two neural networks have the exact same architecture and share the same weights.

Answer 1

在 PyTorch 中实现孪生神经网络就像在不同输入上调用网络函数两次一样简单。

mynet = torch.nn.Sequential(
        nn.Linear(10, 512),
        nn.ReLU(),
        nn.Linear(512, 2))
...
output1 = mynet(input1)
output2 = mynet(input2)
...
loss.backward()

调用 loss.backwad() 时，PyTorch 会自动对两次调用 mynet 的梯度求和。

您可以找到一个完整的示例 here 。