ViMU is a benchmark dataset for video metaphor understanding released by the National University of Singapore in 2026. Related research papers include... ViMU: Benchmarking Video Metaphorical UnderstandingThe aim is to evaluate the ability of multimodal large models to understand the deep semantic meaning of video metaphors.

Mission direction:

• Open-ended interpretation (OE): Evaluating the model's ability to freely interpret the implicit semantics and social meaning of videos.
• Evidence Basis (EG): Evaluating the model's ability to locate video evidence and predict multi-labeled data.
• Rhetorical Mechanism Recognition (RM): Identifying rhetorical expression mechanisms in videos.
• Social Value Signal Recognition (SV): Analyzing Social Values and Implicit Attitude Expressions in Videos

Citation

@article{li2026vimu,
title={ViMU: Benchmarking Video Metaphorical Understanding},
author={Li, Qi and Wang, Xinchao},
journal={arXiv preprint arXiv:2605.14607},
year={2026}
}