If should I resize an image to a square

[xun-dao]

when I use torchserve, I must change my image to a square like(640 * 640), but if I use my model directly, I can inference image like (640* 516) or (640 * any size).

[Loayaltal]

Same issue here using torchscript model in a MAR file