Dynamic Resolution Routing for Efficient Egocentric Grounding

2026-08-02

AI Summary

This work addresses the high computational cost of processing high-resolution visual inputs in egocentric visual grounding and the unreliability of existing compression methods in selecting relevant regions. To this end, the authors propose SmartRes, a framework that introduces a dynamic resolution routing mechanism in pixel space: it first captures global context at low resolution and then employs a lightweight router to selectively activate high-resolution patches in target regions, constructing an order-preserving visual sequence. A novel routing objective with boundary regularization is designed to enhance logit separation between foreground and background, thereby improving recall for small targets. Evaluated on Ego4D and EgoIntention, SmartRes retains 86.4% of full-resolution performance using only 33% of the visual tokens, achieving 1.66× faster inference than the current best compression method while delivering higher accuracy.

Abstract

Egocentric visual grounding requires high-resolution inputs to localize small objects. However, scaling Multimodal Large Language Models to this domain is constrained by the excessive cost of visual token processing. We identify that current efficient strategies based on token reduction are unreliable for selecting object-centric spatial evidence. To overcome this, we propose SmartRes, a framework that performs efficiency optimization in the pixel space via dynamic resolution routing. SmartRes first encodes a low-resolution view for global context and uses a lightweight router to activate high-resolution patches in object-centric regions and constructs an order-preserving visual sequence. To further enable robust routing under severe foreground-background imbalance, we introduce a margin-regularized routing objective that increases foreground-background logit separation and improves foreground recall. Experiments on Ego4D and EgoIntention show that SmartRes reduces visual tokens by up to 67% while retaining 86.4% of full-resolution performance, and achieves up to 1.66X faster inference than state-of-the-art token reduction methods with higher accuracy. Furthermore, strong performance on small object grounding indicates the effectiveness of SmartRes towards egocentric applications. Code will be publicly available.

Problem

• egocentric grounding
• visual token efficiency
• high-resolution input
• small object localization
• multimodal large language models

Innovation

• dynamic resolution routing
• egocentric grounding
• visual token efficiency
• margin-regularized routing
• SmartRes