MTPaperBananaBench is a new benchmark for multi-turn human-in-the-loop scientific diagram generation, featuring 292 images and 3,518 user requirements. It includes a user simulator to scale benchmarking by identifying unsatisfied requirements in automated workflows.
HOW THIS AFFECTS YOU
●
researcherYou can use this to evaluate how well your diagram generation models handle iterative user feedback.
●
designerThis focuses on the interactive, multi-turn nature of creative workflows.