VisNec: Measuring and Leveraging Visual Necessity for Multimodal Instruction Tuning

Open paper
Benchmark Model Rank Results
object-hallucination-evaluation-on-popeVisNec (LLaVA-v1.5-7B, LLaVA-665K 15%)#25F1: 86.0
visual-question-answering-on-mm-vetVisNec (LLaVA-v1.5-7B, LLaVA-665K 15%)#162GPT-4 score: 32.1

Rank counts only results with a code link.