WeMMU: Enhanced Bridging of Vision-Language Models and Diffusion Models via Noisy Query Tokens

TL;DR

This paper introduces Noisy Query Tokens and a VAE branch to improve the integration of Vision-Language Models with Diffusion Models, enabling better task generalization and continual learning in multimodal applications.

Contribution

We propose Noisy Query Tokens and a VAE-based method to enhance task generalization and continual learning in vision-language and diffusion model integration.

Findings

Mitigates generalization collapse in multimodal models

Enables stable continual learning across diverse tasks

Improves fine-grained image detail recovery

Abstract

Recent progress in multimodal large language models (MLLMs) has highlighted the challenge of efficiently bridging pre-trained Vision-Language Models (VLMs) with Diffusion Models. While methods using a fixed number of learnable query tokens offer computational efficiency, they suffer from task generalization collapse, failing to adapt to new tasks that are distant from their pre-training tasks. To overcome this, we propose Noisy Query Tokens, which learn a distributed representation space between the VLM and Diffusion Model via end-to-end optimization, enhancing continual learning. Additionally, we introduce a VAE branch with linear projection to recover fine-grained image details. Experimental results confirm our approach mitigates generalization collapse and enables stable continual learning across diverse tasks.