Deep Visual-Semantic Alignments for Generating Image Descriptionscs.stanford.edu14 pointskercker12 years ago