VISion On Request: Enhanced VLLM efficiency with sparse, dynamically selected, vision-language interactions

TL;DR

VISOR introduces a dynamic, sparse attention mechanism for vision-language models that reduces inference costs without losing visual information, enabling efficient high-resolution reasoning.

Contribution

It proposes a novel sparse, dynamic attention approach that improves efficiency of LVLMs without visual token reduction, maintaining performance on complex tasks.

Findings

Significantly reduces computational cost.

Matches or exceeds state-of-the-art performance.

Excels in tasks requiring detailed visual understanding.

Abstract

Existing approaches for improving the efficiency of Large Vision-Language Models (LVLMs) are largely based on the concept of visual token reduction. This approach, however, creates an information bottleneck that impairs performance, especially on challenging tasks that require fine-grained understanding and reasoning. In this work, we challenge this paradigm by introducing VISion On Request (VISOR), a method that reduces inference cost without discarding visual information. Instead of compressing the image, VISOR improves efficiency by sparsifying the interaction between image and text tokens. Specifically, the language model attends to the full set of high-resolution visual tokens through a small, strategically placed set of attention layers: general visual context is provided by efficient cross-attention between text-image, while a few well-placed and dynamically selected self-attention layers refine the visual representations themselves, enabling complex, high-resolution reasoning when needed. Based on this principle, we first train a single universal network on a range of computational budgets by varying the number of self-attention layers, and then introduce a lightweight policy mechanism that dynamically allocates visual computation based on per-sample complexity. Extensive experiments show that VISOR drastically reduces computational cost while matching or exceeding state-of-the-art results across a diverse suite of benchmarks, and excels in challenging tasks that require detailed visual understanding.