Load pretrained model


I trained VGG16 network for CIFAR10 classification.

I stored this network using torch.save().

And now, I make modified convolution operations and make same architecture like VGG16 with this modified convolution architecture, called modified VGG16. It means Conv2d is replaced by modified convolution operation.

I wanna use pretrained parameters but using modified convolution.

So, I used load_state_dict(torch.load(PATH), strict=False).

However, I got this error messange.

AttributeError: ‘VGG’ object has no attribute ‘copy’

How can I transfer pretrained parameters to modified VGG16??