Analyzing Fine-Grained Alignment and Enhancing Vision Understanding in Multimodal Language Models

TL;DR

This paper investigates how vision embeddings are aligned with language models in multimodal systems, revealing the role of projectors and proposing patch-aligned training to improve visual understanding and downstream task performance.

Contribution

It introduces patch-aligned training to enhance patch-level alignment and compression in multimodal models, leading to better captioning and task performance.

Findings

Improved patch-level alignment with patch-aligned training

Enhanced caption quality and visual understanding

Significant performance gains on grounding and QA tasks

Abstract

Achieving better alignment between vision embeddings and Large Language Models (LLMs) is crucial for enhancing the abilities of Multimodal LLMs (MLLMs), particularly for recent models that rely on powerful pretrained vision encoders and LLMs. A common approach to connect the pretrained vision encoder and LLM is through a projector applied after the vision encoder. However, the projector is often trained to enable the LLM to generate captions, and hence the mechanism by which LLMs understand each vision token remains unclear. In this work, we first investigate the role of the projector in compressing vision embeddings and aligning them with word embeddings. We show that the projector significantly compresses visual information, removing redundant details while preserving essential elements necessary for the LLM to understand visual content. We then examine patch-level alignment -- the alignment between each vision patch and its corresponding semantic words -- and propose a *multi-semantic alignment hypothesis*. Our analysis indicates that the projector trained by caption loss improves patch-level alignment but only to a limited extent, resulting in weak and coarse alignment. To address this issue, we propose *patch-aligned training* to efficiently enhance patch-level alignment. Our experiments show that patch-aligned training (1) achieves stronger compression capability and improved patch-level alignment, enabling the MLLM to generate higher-quality captions, (2) improves the MLLM's performance by 16% on referring expression grounding tasks, 4% on question-answering tasks, and 3% on modern instruction-following benchmarks when using the same supervised fine-tuning (SFT) setting. The proposed method can be easily extended to other multimodal models.