Doing intra-frame prediction backwards (and a few comments)