ActiveScope: Actively Seeking and Correcting Perception for MLLMs

2026-06-23

AI Summary

This work addresses the limitations of existing training-free multimodal large language models in achieving fine-grained, multi-object localization within high-resolution images, where performance is often degraded by distractors and undermined by two fundamental failure modes: context dominance and semantic bias. To overcome these challenges, the authors propose a training-free, active search and self-correction framework that iteratively refines perception results through semantic anchor localization (SAL) and interference suppression refinement (ISR), guided by attention mechanisms. This approach significantly enhances multi-object localization accuracy, achieving a state-of-the-art 96.34% accuracy on the V* Bench benchmark and substantially outperforming current training-free methods.

Abstract

Multimodal Large Language Models (MLLMs) have demonstrated impressive vision-language understanding, yet still struggle with fine-grained perception in high-resolution images. While existing training-free methods typically rely on attention-based localization or coarse-to-fine search, they are often misled by distractors and fail to locate multiple targets. Our investigation attributes these failures to Contextual Dominance, where salient distractors overwhelm target attention and cause inaccurate localization, and Semantic Bias, where global semantics cause the model to fixate on the most salient concept, resulting in incomplete localization in multi-object scenarios. Built on these insights, we propose ActiveScope, a training-free framework that enhances MLLMs by actively seeking and correcting perception. ActiveScope features two modules. The Semantic Anchor Localization (SAL) utilizes fine-grained semantic anchors to independently localize key targets, thereby mitigating semantic bias. The Interference-Suppressed Refinement (ISR) refines localization by suppressing attention on salient distractions to overcome contextual dominance. Extensive experiments on high-resolution image understanding benchmarks demonstrate that ActiveScope outperforms existing training-free methods (e.g., 96.34 percent accuracy on $V^{*}$ Bench), validating the superiority of the active search and self-correction paradigm. Our code is available at https://github.com/jasmine-ww/ActiveScope.

Problem

• Multimodal Large Language Models
• fine-grained perception
• Contextual Dominance
• Semantic Bias
• high-resolution image understanding

Innovation

• ActiveScope
• Semantic Anchor Localization
• Interference-Suppressed Refinement
• Contextual Dominance
• Semantic Bias