Hello, thank you for your answer last time. My code ability is relatively poor, and some parts are not very clear. Could you please explain it to me
This is a two-dimensional convolution method. I think you choose to use nn.Conv2d or nn. functional.Conv2d according to these two parameters self.weight.fast,self.bias.fast , but I don't understand what these two parameters mean, According to the function description, it seems to be the tensor of filter,tensor of f bias,but i dont konw why do it.The code I usually see is to directly use nn.conv2d
I understand the meaning of this method roughly. It is based on whether the name of the network parameter contains gamma, beta. the parameter is divided into model and FW. What I don't understand is why I do this. you saved these data. It seems that you used these data in the later training.
@hytseng0509
Hello, thank you for your answer last time. My code ability is relatively poor, and some parts are not very clear. Could you please explain it to me
This is a two-dimensional convolution method. I think you choose to use nn.Conv2d or nn. functional.Conv2d according to these two parameters self.weight.fast,self.bias.fast , but I don't understand what these two parameters mean, According to the function description, it seems to be the tensor of filter,tensor of f bias,but i dont konw why do it.The code I usually see is to directly use nn.conv2d