Visual Metaphor

Visual metaphor research explores how meaning is conveyed implicitly through images, often in conjunction with text, focusing on understanding and generating these metaphorical representations. Current research utilizes large language models (LLMs) coupled with diffusion models or vision-language (VL) models to create and interpret visual metaphors, often leveraging newly created datasets of memes, videos, and drawings to benchmark performance. This work is significant for advancing AI's ability to understand nuanced human communication and has implications for applications in education, therapy, and the detection of propaganda in online media.