Normalized and Geometry-Aware Self-Attention Network for Image Captioning | IEEE Conference Publication | IEEE Xplore