Study examines how VLMs handle novel visual references

Arxiv CS.CV (computer vision)

The paper introduces a framework to study how vision-language models map novel visual concepts to language, especially when they contradict prior knowledge. Experiments show VLMs exhibit human-like patterns but struggle with conflicting references.

vision-language-models

Arxiv CS.CV (computer vision)·Ada Defne T\"urGaurav KamathJoyce ChaiSiva ReddyBenno Krojer

6 days ago