MMFuser: Multimodal Multi-Layer Feature Fuser for Fine-Grained Vision-Language Understanding

TL;DR

MMFuser enhances vision-language understanding by efficiently integrating multi-layer features from Vision Transformers, capturing fine-grained details without added redundancy, leading to improved benchmark performance.

Contribution

Introduces a multi-layer feature fuser that dynamically combines deep and shallow features from ViTs, improving visual detail representation in MLLMs.

Findings

Significant performance improvements on benchmarks.

Efficient integration of multi-layer features.

Lightweight alternative to multi-encoder methods.

Abstract

Despite significant advancements in Multimodal Large Language Models (MLLMs) for understanding complex human intentions through cross-modal interactions, capturing intricate image details remains challenging. Previous methods integrating multiple vision encoders to enhance visual detail introduce redundancy and computational overhead. We observe that most MLLMs utilize only the last-layer feature map of the vision encoder for visual representation, neglecting the rich fine-grained information in shallow feature maps. To address this issue, we propose \modelname, a simple yet effective multi-layer feature fuser that efficiently integrates deep and shallow features from Vision Transformers (ViTs). Specifically, it leverages semantically aligned deep features as queries to dynamically extract missing details from shallow features, thus preserving semantic alignment while enriching the representation with fine-grained information. Applied to the LLaVA-1.5 model, \modelname~achieves significant improvements in visual representation and benchmark performance, providing a more flexible and lightweight solution compared to multi-encoder ensemble methods. The code and model have been released at https://github.com/yuecao0119/MMFuser.