LLaVA-UHD v4: What Makes Efficient Visual Encoding in MLLMs?

TL;DR

LLaVA-UHD v4 introduces a slice-based encoding and intra-ViT early compression to significantly reduce visual-encoding FLOPs in high-resolution MLLMs while maintaining or improving performance.

Contribution

It proposes a novel intra-ViT early compression method combined with slice-based encoding, enhancing efficiency in high-resolution visual encoding for MLLMs.

Findings

Reduces visual-encoding FLOPs by 55.8%

Outperforms global encoding in benchmarks

Maintains or surpasses baseline performance

Abstract

Visual encoding constitutes a major computational bottleneck in Multimodal Large Language Models (MLLMs), especially for high-resolution image inputs. The prevailing practice typically adopts global encoding followed by post-ViT compression. Global encoding produces massive token sequences, while post-ViT compression incurs the full quadratic attention cost of the ViT before any token reduction takes place. In this work, we revisit this convention along two dimensions: the encoding strategy and visual token compression. First, controlled experiments show that slice-based encoding outperforms global encoding across benchmarks, suggesting that preserving local details through sliced views can be more beneficial than applying global attention for fine-grained perception. Second, we introduce intra-ViT early compression, which reduces tokens in shallow ViT layers and substantially lowers visual-encoding FLOPs while preserving downstream performance. By integrating intra-ViT compression into the slice-based encoding framework, we present LLaVA-UHD v4, an efficient and compute-controllable visual encoding scheme tailored for high-resolution inputs. Across a diverse set of benchmarks covering document understanding, OCR, and general VQA, LLaVA-UHD v4 reduces visual-encoding FLOPs by 55.8% while matching or even surpassing baseline performance. These results suggest that visual-encoding efficiency can be substantially improved without sacrificing downstream performance, providing a practical design direction for efficient high-resolution MLLMs. All model weights and code will be publicly released to support further research.