Torch model forward with a diferent image size

Question

I am testing some well known models for computer vision: UNet, FC-DenseNet103, this implementation I train them with 224x224 randomly cropped patches and do the same on the vali