RBench-V
Emerging7papers using it
2025first seen
RBench-V is a benchmark containing 803 questions designed to evaluate the visual reasoning abilities of multi-modal models through tasks that require generating and manipulating images as part of the reasoning process.
Papers using RBench-V (7)
- SpecEyes: Accelerating Agentic Multimodal LLMs via Speculative Perception and PlanningActiveScope: Actively Seeking and Correcting Perception for MLLMsHART: High-Resolution Annotation-Free Reasoning Technique through a Closed-loop FrameworkGuiding the Inner Eye: A Framework for Hierarchical and Flexible Visual Grounded ReasoningGrounded Reinforcement Learning For Visual ReasoningGam-agent: Game-theoretic And Uncertainty-aware Collaboration For Complex Visual ReasoningRBench-V: A Primary Assessment for Visual Reasoning Models with Multi-modal Outputs