State Beyond Appearance: Diagnosing and Improving State Consistency in Dial-Based Measurement Reading

2026-04-29

AI Summary

This work addresses the sensitivity of multimodal large language models to viewpoint and illumination variations in dial reading tasks, which undermines their ability to maintain state consistency due to neglecting the intrinsic geometric structure of dial states. The study is the first to identify this issue and introduces TriSCA, a three-level state-consistent alignment framework that explicitly models state consistency through state-distance-aware representation alignment, metadata-guided observation-to-state supervision, and state-aware objective optimization. Evaluated on both controlled and real-world benchmarks involving clocks and instruments, TriSCA substantially improves reading accuracy and robustness, overcoming the limitations of existing approaches that rely solely on superficial visual cues.

Abstract

Multimodal large language models (MLLMs) have achieved impressive progress on general multimodal tasks, yet they remain brittle on dial-based measurement reading. In this paper, we study this problem through controlled benchmarks and feature-space probing, and show that current MLLMs not only achieve unsatisfactory accuracy on dial-based readout, but also suffer sharp performance drops under viewpoint and illumination changes even when the underlying dial state remains fixed. Our probing analysis further reveals that same-state samples under appearance variation are not consistently clustered, while neighboring states fail to preserve the local structure implied by continuous dial values. These findings suggest that existing MLLMs largely ignore the intrinsic state geometry of dial measurement tasks and instead rely on superficial appearance cues. Motivated by this diagnosis, we propose TriSCA, a tri-level state-consistent alignment framework for dial-based measurement reading. Specifically, TriSCA consists of state-distance-aware representation alignment, metadata-grounded observation-to-state supervision, and state-aware objective alignment. Extensive ablation studies and evaluation experiments on controlled clock and gauge benchmarks, together with evaluation on an external real-world benchmark, demonstrate the effectiveness of our method.

Problem

• dial-based measurement reading
• state consistency
• multimodal large language models
• appearance variation
• state geometry

Innovation

• state consistency
• dial-based measurement reading
• multimodal large language models
• representation alignment
• geometric structure