It seems all pictures in MVTec-AD dataset are in square shape (1000×1000, 1024×1024, etc.). I wonder is it necessary or required to be?
When making our own dataset in MVTec format by converting existing dataset in other format like COCO format, do pictures have to be sliced into square shape? Or leaving them with random size (1920×1080, 640×480, etc.) is okay?
Or this is depending on what AI anomaly-detection framework/lib we used?