如何对孪生网络特征进行可视化

[IceHowe]

如何对孪生网络特征进行可视化，输入是两张图像，backbone经过了两次，每张图像经过一次，例如：

self.backbone = resnet18()

forward(x1,x2):
    fx1 = self.backbone(x1)
    fx2 = self.backbone(x2)

[frgfm]

Hello @IceHowe 👋

I'm sorry, but I had to use Google Translate to understand your request so my understanding might be limited :/

Now regarding your question about visualizing a twin network:

• I'm not sure how, in your example, this is a twin network
• with your snippet, I suggest using the CAM extractor twice: after the first forward, and after the last

I'm pretty sure I didn't get your actual request, so feel free to specify 😅
Would you mind translating to English for your next message please? (or if you prefer French, German, Spanish)
Cheers ✌️

[vegeckn]

I have the same problem with you. And the method throws out an error: forward() missing 1 required positional argument: 'x2'. I do not know how to solve this problem. Do you solve this ?