Could I convert the image into a tensor that could then be reconstruced into an image again?