From Model Uncertainty to Human Attention: Localization-Aware Visual Cues for Scalable Annotation Review

2026-05-12

AI Summary

This work addresses a critical limitation in current AI-assisted annotation pipelines, which predominantly rely on classification confidence and often fail to signal spatial localization errors in bounding boxes, leading human reviewers to overlook subtle misalignments and compromising annotation quality. To overcome this, the study introduces localization uncertainty as a key cue for human–AI collaboration. By quantifying the model’s spatial uncertainty and integrating it into a dedicated visualization interface, the approach directs annotators’ attention toward regions of high uncertainty. In a controlled experiment involving 120 participants, this uncertainty-guided method significantly improved both annotation accuracy and efficiency, demonstrating that leveraging localization uncertainty effectively optimizes human reviewers’ attention allocation and transcends the constraints of conventional confidence-based paradigms.

Abstract

High-quality labeled data is essential for training robust machine learning models, yet obtaining annotations at scale remains expensive. AI-assisted annotation has therefore become standard in large-scale labeling workflows. However, in tasks where model predictions carry two independent components, a class label and spatial boundaries, a model may classify an object with high confidence while mislocalizing it. Existing AI-assisted workflows offer annotators no signal about where spatial errors are most likely. Without such guidance, humans may systematically underinspect subtly misplaced boxes. We address this by studying the effect of visualizing spatial uncertainty via a purpose-built interface. In a controlled study with 120 participants, those receiving uncertainty cues achieve higher label quality while being faster overall. A box-level analysis confirms that the cues redirect annotator effort toward high-uncertainty predictions and away from well-localized boxes. These findings establish localization uncertainty as a lever to improve human-in-the-loop annotation. Code is available at https://mos-ks.github.io/MUHA/.

Problem

• model uncertainty
• localization error
• AI-assisted annotation
• human attention
• scalable annotation

Innovation

• localization uncertainty
• AI-assisted annotation
• human-in-the-loop
• visual cues
• spatial boundaries