What is the size of the in features Resnet 50

vision
1

Hi guys, I need to know the size of the “In Features” (before Fully connected Layer) of resNet 50.
Is it 7x7x2048? Is the right size?
And if I want this “Image” this code:

class ResNetMultiHead(nn.Moule):
    def __init__(self):
        super(ResNetMultiHead, self).__init__()

        self.model = torchvison.models.resnet50()
        self.model.fc = nn.identity()

    def forward(self, x):

        head1 = self.model(x)
        head2 = head1
        head3 = head1

        return head1, head2, head3

Is it right?

So each head is 7x7x2048?

2

Channel should be 2048 if directly using the last stage in resnet-50, but height and width should be correlated with your input image’s resolution. The stride between input and the last feature map is 32, which means if your input size is 224 * 224 * 3, then the output (head1) should be 7 * 7 * 2048

3

Ok , I make the exact calculation.
Thanks for your reply