Interpreting Art: Visual Question Answering in Paintings

It is challenging to respond to queries about paintings because it requires knowledge of both the visual details depicted in the image and the background information learned by studying art history. Using a dataset of paintings and comments from an existing art dataset, we employed state-of-the-art question generation techniques to automatically construct question-answer (QA) pairs in this study. Our study presents the outcomes of this review, along with a structured workflow encompassing phases for Visual Question Answering. In each phase, we explore and discuss the techniques found in literature. Additionally, we examine evaluation techniques employed to gauge the quality of generated Answers. Lastly, we identify areas that need more investigation to improve the current state of knowledge on visual question answering in art.