Image width and height are not equal

[SaulZhang]

It seems that the code can't calculate the metric on the images with unequal width and height, can it be expanded to calculate images of various sizes?

[richzhang]

I think it should work out of the box

model = models.PerceptualLoss(model='net-lin', net='vgg', use_gpu=False, spatial=False)

H,W = 160, 64
dummy = torch.zeros(1,3,H,W)
print(model(dummy, dummy).item())

[SaulZhang]

Thanks for your answer, it does work in the example you provided. After my testing as the test_network.py script settings, I found that this problem is mainly caused by the setting of {net/spatial} parameter.

model = models.PerceptualLoss(model='net-lin', net='alex', use_gpu=False, spatial=True)
H,W = 160, 64
dummy = torch.zeros(1,3,H,W)

the error information as follow:
RuntimeError: The size of tensor a (61) must match the size of tensor b (58) at non-singleton dimension 3

[richzhang]

Got it. I pushed a change to do resizing in both dimensions. Let me know if it works. Keep in mind the features in each layer will not necessarily be well-aligned in these cases

[khitish19]

The issue still persists for me. I was trying compute_dists_pair.py and got the same error when the image height and width were different but it all worked fine when I resize them to have same height and width.

The error looks like this:
RuntimeError: The size of tensor a (13) must match the size of tensor b (15) at non-singleton dimension 3

[be42day]

The two images must be in the same size