Output batch size changes when input image size is modified

[venki-lfc]

Hi,
First of all, thanks for a great paper! Its quite interesting to see how you try to separate ID and OOD classes.

When I try to apply this concept for my data, I am facing some issues during the WRN model creation.
My input transform looks like the following

transform = T.Compose([
    T.Resize(320, interpolation=T.InterpolationMode.BICUBIC),
    T.CenterCrop(300),
    T.ToTensor(),
    T.Normalize(mean=[0.485, 0.456, 0.406],
                std=[0.229, 0.224, 0.225]), ])

As you can see my images are of the size 300 (In contrast to 32 that you guys do for CIFAR-10).

Now when I feed these images to the model, the output batch size is totally different from what I feed in. For example: With the above transformation and batch size of 2, my output size is (162, num_classes).
But if the transformation size is 32, I get the right output (2, num_classes). May I know what is happening here?

Another doubt: Can your OECC concept be applied to any other model? EfficientNet for example?

Many thanks!
Venki