What is the correct tensor?

vision
1

Hi guys i’ve extracted the features from resnet50 before the layer of Average pooling.
I achieve a tensor of 1x100352, but i have to respect the structure of the image, so what is the exact shape of the tensor? 2048x7x7 or 7x7x2048?

2

In PyTorch the default is channels first, so the Tensor should be of shape [B, C, H, W] -> [1, 2048, 7, 7]