In all of the public image datasets I have seen the subject in the image has been neatly snipped around leaving only plain background. Is this necessary in making a training da
