Test-time Scaling over Perception: Resolving the Grounding Paradox in Thinking with Images

TL;DR

This paper introduces TTSP, a test-time scaling framework for perception in multimodal models, improving fine-grained visual reasoning by iteratively exploring and refining visual evidence.

Contribution

The paper proposes a novel test-time perception scaling method that enhances multimodal reasoning by generating and validating multiple visual exploration traces.

Findings

TTSP outperforms strong baselines on high-resolution and general multimodal benchmarks.

TTSP demonstrates scalability and token efficiency in visual reasoning tasks.

Scaling perception at test time improves robustness under perceptual uncertainty.

Abstract

Recent multimodal large language models (MLLMs) have begun to support Thinking with Images by invoking visual tools such as zooming and cropping during inference. Yet these systems remain brittle in fine-grained visual reasoning because they must decide where to look before they have access to the evidence needed to make that decision correctly. We identify this circular dependency as the Grounding Paradox. To address it, we propose Test-Time Scaling over Perception (TTSP), a framework that treats perception itself as a scalable inference process. TTSP generates multiple exploratory perception traces, filters unreliable traces using entropy-based confidence estimation, distills validated observations into structured knowledge, and iteratively refines subsequent exploration toward unresolved uncertainty. Extensive experiments on high-resolution and general multimodal reasoning benchmarks show that TTSP consistently outperforms strong baselines across backbone sizes, while also exhibiting favorable scalability and token efficiency. Our results suggest that scaling perception at test time is a promising direction for robust multimodal reasoning under perceptual uncertainty.