Exploiting Multiple Sequence Lengths in Fast End to End Training for Image Captioning | IEEE Conference Publication | IEEE Xplore