Multi-level Visual Fusion Networks for Image Captioning | IEEE Conference Publication | IEEE Xplore