MG-LLaVA: Towards Multi-Granularity Visual Instruction Tuning

Open paper

Benchmark	Model	Rank	Results
visual-question-answering-on-mm-vet	MG-LLaVA(34B)	#53	GPT-4 score: 48.5