Unleashing Perception-Time Scaling to Multimodal Reasoning Models

2025-10-09

AI Summary

Current large vision-language models (LVLMs) suffer from limited accuracy in visual perception tasks and exhibit marginal gains from inference-time scaling, primarily because their “fast perception” paradigm neglects the temporal nature of perception and the utility of intermediate representations. To address this, we propose Perception Temporal Scaling (PTS), a novel paradigm that introduces inference-time scaling to multimodal visual understanding for the first time. PTS explicitly decomposes perception into sequential subtasks and generates token-rich intermediate representations, enabling joint optimization of perception and reasoning. Our reinforcement learning–based PTS framework achieves a substantial improvement in high-precision rate—from 8.0% to 64.7%—on our newly constructed visual estimation benchmark, DisTANCE. Furthermore, PTS demonstrates strong generalization across domains and robust fine-grained image token attention on diverse real-world benchmarks.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Recent advances in inference-time scaling, particularly those leveraging reinforcement learning with verifiable rewards, have substantially enhanced the reasoning capabilities of Large Vision-Language Models (LVLMs). Inspired by this success, similar strategies have been applied to multimodal reasoning, yet their impact on visual perception remains unclear. To investigate this gap, we introduce DisTANCE, a perception-centric benchmark for visual estimation tasks. Evaluation results show that LVLMs exhibit limited estimation precision, and inference-time scaling offers only marginal gains. We attribute this to the fast perception paradigm of current LVLMs, where visual understanding is treated as a one-shot output without modeling the underlying perceptual process. To address this, we propose Perception-Time Scaling (PTS), a novel paradigm that encourages token-rich perception and decomposes complex perception problems into intermediate tractable sub-problems, thereby enabling perception to align with and benefit from inference-time scaling. Combined with reinforcement learning techniques, PTS significantly improves perception accuracy, raising high-precision performance on DisTANCE from 8.0% to 64.7%, and generalizes well to out-of-domain tasks. Surprisingly, even though PTS data are purely synthetic, combining them with math reasoning data yields consistent gains in both reasoning and real-world perception benchmarks. Further analysis reveals that PTS introduces more perception-related tokens and increases the model's attention to image tokens. Our code and data will be publicly released.

Problem

• Limited visual estimation precision in multimodal reasoning models
• Fast perception paradigm treats vision as one-shot output
• Current inference-time scaling offers marginal perception gains

Innovation

• Introduces Perception-Time Scaling for multimodal reasoning
• Decomposes perception into intermediate sub-problems
• Combines synthetic data with reinforcement learning techniques