UI-Zoomer: Uncertainty-Driven Adaptive Zoom-In for GUI Grounding

TL;DR

UI-Zoomer is an adaptive, uncertainty-driven zoom-in framework for GUI grounding that improves localization accuracy without additional training by selectively cropping based on prediction confidence.

Contribution

It introduces a training-free, uncertainty-based adaptive zoom-in method that dynamically adjusts crop sizes for better GUI element localization.

Findings

Achieves up to +13.4% improvement on ScreenSpot-Pro

Demonstrates consistent gains across multiple datasets and models

Operates without additional training, reducing complexity

Abstract

GUI grounding, which localizes interface elements from screenshots given natural language queries, remains challenging for small icons and dense layouts. Test-time zoom-in methods improve localization by cropping and re-running inference at higher resolution, but apply cropping uniformly across all instances with fixed crop sizes, ignoring whether the model is actually uncertain on each case. We propose \textbf{UI-Zoomer}, a training-free adaptive zoom-in framework that treats both the trigger and scale of zoom-in as a prediction uncertainty quantification problem. A confidence-aware gate fuses spatial consensus among stochastic candidates with token-level generation confidence to selectively trigger zoom-in only when localization is uncertain. When triggered, an uncertainty-driven crop sizing module decomposes prediction variance into inter-sample positional spread and intra-sample box extent, deriving a per-instance crop radius via the law of total variance. Extensive experiments on ScreenSpot-Pro, UI-Vision, and ScreenSpot-v2 demonstrate consistent improvements over strong baselines across multiple model architectures, achieving gains of up to +13.4\%, +10.3\%, and +4.2\% respectively, with no additional training required.