HRBench-4K

HRBench-4K is a step-wise grounded reasoning dataset that contains visual evidence with bounding boxes and region-specific semantic descriptions, used to evaluate the performance of multimodal large language models in spatial-semantic reasoning tasks.

Papers using HRBench-4K (3)

Thinking With Images Via Self-calling Agent2025Retrieve, Integrate, and Synthesize: Spatial-Semantic Grounded Latent Visual Reasoning2026Small Drafts, Big Verdict: Information-Intensive Visual Reasoning via Speculation2025