A method for converting images to a unified image format for artificial intelligence deep training comprises acquiring a plurality of images and annotating the plurality of images. An annotation site of the plurality of images is determined to be quadrilateral or polygonal. The format of the annotated images is converted according to a first conversion rule when the annotation site of the annotated training images or the annotated verification images is found to be quadrilateral and the format of the annotated images is converted according to a second conversion rule when the annotation site of the annotated training images or the annotated verification images is found to be not a quadrilateral. A device employing the method is also disclosed.