GranFact

New benchmark and method improve fine-grained image description in LLMs

Researchers have identified a challenge in multimodal large language models (MLLMs) where generating fine-grained visual descriptions is more prone to errors than coarse-grained ones. To address this, they developed Gra…