SiMing-Bench: Evaluating Procedural Correctness from Continuous Interactions in Clinical Skill Videos
🤖 AI Summary
Existing video evaluation benchmarks struggle to assess the ability of multimodal large language models (MLLMs) to dynamically reason about interaction-driven procedural state changes in clinical skills videos. To address this gap, this work introduces SiMing-Score, the first benchmark specifically designed for end-to-end clinical procedure evaluation. Built upon real-world videos annotated by physicians under double-blind conditions and aligned with standardized scoring rubrics, SiMing-Score proposes a novel evaluation paradigm centered on “procedural correctness” through dynamic state updates. Step-level alignment analysis reveals that while prevailing MLLMs exhibit moderate correlation with expert judgments at the global procedural level, they significantly underperform in fine-grained assessment of step-wise procedural correctness, highlighting their limited capacity to model state evolution triggered by continuous human–environment interactions.
Application Category
📝 Abstract
Current video benchmarks for multimodal large language models (MLLMs) focus on event recognition, temporal ordering, and long-context recall, but overlook a harder capability required for expert procedural judgment: tracking how ongoing interactions update the procedural state and thereby determine the correctness of later actions. We introduce SiMing-Bench, the first benchmark for evaluating this capability from full-length clinical skill videos. It targets rubric-grounded process-level judgment of whether interaction-driven state updates preserve procedural correctness across an entire workflow. SiMing-Bench is instantiated with SiMing-Score, a physician-annotated dataset of real clinical skill examination videos spanning cardiopulmonary resuscitation, automated external defibrillator operation, and bag-mask ventilation, each paired with a standardized step-wise rubric and dual-expert labels. Across diverse open- and closed-source MLLMs, we observe consistently weak agreement with physician judgments. Moreover, weak performance on rubric-defined intermediate steps persists even when overall procedure-level correlation appears acceptable, suggesting that coarse global assessment substantially overestimates current models'procedural judgment ability. Additional analyses with binary step judgment and step-aligned clips indicate that the bottleneck is not merely fine-grained scoring or temporal localization, but modeling how continuous interactions update procedural state over time.
Problem
Research questions and friction points this paper is trying to address.
procedural correctness
continuous interactions
clinical skill videos
procedural state
multimodal large language models
Innovation
Methods, ideas, or system contributions that make the work stand out.
procedural correctness
continuous interaction
clinical skill evaluation
multimodal large language models
state-aware judgment
🔎 Similar Papers
No similar papers found.
