AI RESEARCH
LLaVA-UHD v4: What Makes Efficient Visual Encoding in MLLMs?
arXiv CS.CV
•
ArXi:2605.08985v1 Announce Type: new 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.