利用视觉定位中的缩放一致性，免费获得模型置信度信号。
Zoom Consistency: A Free Confidence Signal in Multi-Step Visual Grounding Pipelines
- 用缩放过程中的预测偏差作为几何置信度信号。
- 该信号与预测正确性相关，跨模型、系统均有效（皮尔逊相关系数-0.14至-0.11）。
- 可指导模型路由，提升性能，适合研究多步视觉定位的开发者。
多步缩放定位流程广泛用于GUI定位任务，但其中间输出常被丢弃。我们发现这些中间结果包含一个无需额外计算的有用置信度信号：缩放一致性，即第二步预测与裁剪中心之间的距离。不同于对数概率或标记级不确定性，缩放一致性是共享坐标空间中的几何量，可直接在不同架构的视觉语言模型间比较，无需校准。在理想条件下，该量为第一步空间误差的线性估计器。实验表明，它在两个模型（KV-Ground-8B和Qwen3.5-27B）上均与预测正确性显著相关（AUC=0.60；Spearman rho=-0.14，p<10⁻⁶；rho=-0.11，p=0.0003）。相关性虽小但稳定，适用于不同模型、应用类别和操作系统。作为概念验证，我们利用该信号实现专家与通用模型的路由策略，捕获了两者之间16.5%的最优头距（+0.8%，McNemar p=0.19）。代码已开源。
原文摘要 · Abstract (English)
Multi-step zoom-in pipelines are widely used for GUI grounding, yet the intermediate predictions they produce are typically discarded after coordinate remapping. We observe that these intermediate outputs contain a useful confidence signal for free: zoom consistency, the distance between a model's step-2 prediction and the crop center. Unlike log-probabilities or token-level uncertainty, zoom consistency is a geometric quantity in a shared coordinate space, making it directly comparable across architecturally different VLMs without calibration. We prove this quantity is a linear estimator of step-1 spatial error under idealized conditions (perfect step-2, target within crop) and show it correlates with prediction correctness across two VLMs (AUC = 0.60; Spearman rho = -0.14, p < 10^{-6} for KV-Ground-8B; rho = -0.11, p = 0.0003 for Qwen3.5-27B). The correlation is small but consistent across models, application categories, and operating systems. As a proof-of-concept, we use zoom consistency to route between a specialist and generalist model, capturing 16.5% of the oracle headroom between them (+0.8%, McNemar p = 0.19). Code is available at https://github.com/omxyz/zoom-consistency-routing.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。