Towards Multimodal Understanding via Stable Diffusion as a Task-Aware Feature Extractor

TL;DR

This paper explores using pre-trained text-to-image diffusion models as visual encoders for multimodal tasks, showing they provide rich, instruction-aware features that improve visual question answering and reasoning capabilities.

Contribution

It introduces a novel approach of leveraging diffusion model features for multimodal understanding, addressing limitations of existing encoders like CLIP, and proposes methods to align and fuse these features with language models.

Findings

Diffusion features are rich in semantics and encode strong image-text alignment.

Text conditioning in diffusion models helps focus on relevant image regions.

Fusion of CLIP and diffusion features improves performance on VQA and reasoning tasks.

Abstract

Recent advances in multimodal large language models (MLLMs) have enabled image-based question-answering capabilities. However, a key limitation is the use of CLIP as the visual encoder; while it can capture coarse global information, it often can miss fine-grained details that are relevant to the input query. To address these shortcomings, this work studies whether pre-trained text-to-image diffusion models can serve as instruction-aware visual encoders. Through an analysis of their internal representations, we find diffusion features are both rich in semantics and can encode strong image-text alignment. Moreover, we find that we can leverage text conditioning to focus the model on regions relevant to the input question. We then investigate how to align these features with large language models and uncover a leakage phenomenon, where the LLM can inadvertently recover information from the original diffusion prompt. We analyze the causes of this leakage and propose a mitigation strategy. Based on these insights, we explore a simple fusion strategy that utilizes both CLIP and conditional diffusion features. We evaluate our approach on both general VQA and specialized MLLM benchmarks, demonstrating the promise of diffusion models for visual understanding, particularly in vision-centric tasks that require spatial and compositional reasoning. Our project page can be found https://vatsalag99.github.io/mustafar/.