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.
Abstract (translated)
多模态大型语言模型（MLLM）在各种视觉-语言任务上取得了显著进展，但其视觉感知能力仍然有限。相比之下，人类通过动态扫描和聚焦于场景中的重要区域，在类似“眨眼”的序列过程中高效地处理复杂场景。受到这一策略的启发，我们首先研究了MLLM是否表现出类似的特性。初步分析显示，MLLM在不同层自然关注不同的视觉区域，并且选择性地向重要的标记分配更多计算资源可以增强其视觉感知能力。基于这一见解，我们提出了一种名为Blink的动态视觉标记解析框架，在单次前向传递中模仿了人类启发的过程。具体而言，Blink包括两个模块：基于显著性的扫描和动态令牌解析。它首先根据注意力图估计每一层中视觉令牌的重要性，并通过即插即用令牌超分辨率（TokenSR）模块扩展重要令牌。在下一层中，当这些扩展的令牌失去焦点时，它们会被丢弃。这种动态机制平衡了广泛的探索与精细聚焦之间的关系，从而能够适应性地、高效地增强视觉感知能力。大量的实验验证了Blink的有效性，证明其可以显著提高视觉感知和多模态理解的能力。
URL
https://arxiv.org/abs/2512.10548
