Blink: Dynamic Visual Token Resolution for Enhanced Multimodal Understanding

TL;DR

Blink introduces a human-inspired dynamic visual token resolution method for multimodal models, improving their ability to perceive complex scenes by focusing on salient regions adaptively within a single pass.

Contribution

The paper proposes Blink, a novel framework that dynamically allocates computational resources to salient visual tokens, enhancing multimodal understanding in large language models.

Findings

Blink improves visual perception in multimodal models.

Dynamic token resolution enhances scene understanding.

Experimental results show significant performance gains.

Abstract

Multimodal large language models (MLLMs) have achieved remarkable progress on various vision-language tasks, yet their visual perception remains limited. Humans, in comparison, perceive complex scenes efficiently by dynamically scanning and focusing on salient regions in a sequential "blink-like" process. Motivated by this strategy, we first investigate whether MLLMs exhibit similar behavior. Our pilot analysis reveals that MLLMs naturally attend to different visual regions across layers and that selectively allocating more computation to salient tokens can enhance visual perception. Building on this insight, we propose Blink, a dynamic visual token resolution framework that emulates the human-inspired process within a single forward pass. Specifically, Blink includes two modules: saliency-guided scanning and dynamic token resolution. It first estimates the saliency of visual tokens in each layer based on the attention map, and extends important tokens through a plug-and-play token super-resolution (TokenSR) module. In the next layer, it drops the extended tokens when they lose focus. This dynamic mechanism balances broad exploration and fine-grained focus, thereby enhancing visual perception adaptively and efficiently. Extensive experiments validate Blink, demonstrating its effectiveness in enhancing visual perception and multimodal understanding.