VinVL: Revisiting Visual Representations in Vision-Language Models | IEEE Conference Publication | IEEE Xplore