Random crop "End-to-end optimized image compressio

[Mareeta Mathai]

Hi,
I'm relatively new to the data augmentation methods.I would like to know more the cropping done on the paper "End-to-end optimized image compression".
What I understood is the images are randomly cropped to 256x256 patches and then fed to the training model. I tried to remove the cropping in training function, but couldn't train since the image was no longer 256x256.
How does it work for testing? I found that testing images are not cropped or resized. How does the trained model recognize the new sizes in the testing set?
Please advise.

Thanks,
Mareeta