Dense Connector for MLLMs
Huanjin Yao, Wenhao Wu, Taojiannan Yang, YuXin Song, Mengxi Zhang, Haocheng Feng, Yifan Sun, Zhiheng Li, Wanli Ouyang, Jingdong Wang
TL;DR
Dense Connector tackles the underutilization of visual signals in MLLMs by fusing multi‑layer features from frozen vision encoders into the LLM via three lightweight instantiations (STI, SCI, DCI). An Efficient Dense Connector reduces token counts with minimal accuracy loss, and a training‑free extension enables video understanding. Across multiple vision backbones, image resolutions, training data scales, and LLM sizes (2.7B–70B), Dense Connector yields strong gains and achieves state‑of‑the‑art results on 19 image/video benchmarks, including competitive video performance with a training‑free approach. The work demonstrates the practical viability and wide applicability of dense multi‑layer visual integration for improving cross‑modal reasoning in MLLMs.
Abstract
Do we fully leverage the potential of visual encoder in Multimodal Large Language Models (MLLMs)? The recent outstanding performance of MLLMs in multimodal understanding has garnered broad attention from both academia and industry. In the current MLLM rat race, the focus seems to be predominantly on the linguistic side. We witness the rise of larger and higher-quality instruction datasets, as well as the involvement of larger-sized LLMs. Yet, scant attention has been directed towards the visual signals utilized by MLLMs, often assumed to be the final high-level features extracted by a frozen visual encoder. In this paper, we introduce the Dense Connector - a simple, effective, and plug-and-play vision-language connector that significantly enhances existing MLLMs by leveraging multi-layer visual features, with minimal additional computational overhead. Building on this, we also propose the Efficient Dense Connector, which achieves performance comparable to LLaVA-v1.5 with only 25% of the visual tokens. Furthermore, our model, trained solely on images, showcases remarkable zero-shot capabilities in video understanding as well. Experimental results across various vision encoders, image resolutions, training dataset scales, varying sizes of LLMs (2.7B->70B), and diverse architectures of MLLMs (e.g., LLaVA-v1.5, LLaVA-NeXT and Mini-Gemini) validate the versatility and scalability of our approach, achieving state-of-the-art performance across 19 image and video benchmarks. We hope that this work will provide valuable experience and serve as a basic module for future MLLM development. Code is available at https://github.com/HJYao00/DenseConnector .