Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering
Top-down visual attention mechanisms have been used extensively in image captioning and visual question answering (VQA) to enable deeper image understanding through fine-grained...