Tag
This paper presents a benchmark and a multi-agent system to improve scientific diagram generation through multi-turn human feedback, addressing issues like quality drift and forgetting across revisions.