Efficient Rectified Flow for Image Fusion

2025-09-20

AI Summary

Image fusion requires integrating complementary multimodal information, yet existing diffusion models suffer from prohibitively long inference times and high computational overhead. This paper proposes RFfusion, an efficient single-step diffusion framework built upon Rectified Flow. Its core innovations include: (i) enabling single-step sampling without additional training; (ii) introducing a task-specific variational autoencoder (VAE) that performs cross-modal feature alignment and fusion directly in the latent space; and (iii) adopting a two-stage training strategy to jointly optimize reconstruction fidelity and semantic consistency of fused representations. Evaluated on multiple benchmarks, RFfusion achieves over 100× speedup relative to typical diffusion-based methods while significantly outperforming state-of-the-art approaches in fusion quality. It delivers millisecond-level inference latency without sacrificing fine-grained structural details—marking the first diffusion-based method to unify efficiency and high-fidelity detail preservation.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Multimodal Learning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization

Abstract

Image fusion is a fundamental and important task in computer vision, aiming to combine complementary information from different modalities to fuse images. In recent years, diffusion models have made significant developments in the field of image fusion. However, diffusion models often require complex computations and redundant inference time, which reduces the applicability of these methods. To address this issue, we propose RFfusion, an efficient one-step diffusion model for image fusion based on Rectified Flow. We incorporate Rectified Flow into the image fusion task to straighten the sampling path in the diffusion model, achieving one-step sampling without the need for additional training, while still maintaining high-quality fusion results. Furthermore, we propose a task-specific variational autoencoder (VAE) architecture tailored for image fusion, where the fusion operation is embedded within the latent space to further reduce computational complexity. To address the inherent discrepancy between conventional reconstruction-oriented VAE objectives and the requirements of image fusion, we introduce a two-stage training strategy. This approach facilitates the effective learning and integration of complementary information from multi-modal source images, thereby enabling the model to retain fine-grained structural details while significantly enhancing inference efficiency. Extensive experiments demonstrate that our method outperforms other state-of-the-art methods in terms of both inference speed and fusion quality. Code is available at https://github.com/zirui0625/RFfusion.

Problem

• Diffusion models for image fusion require complex computations and slow inference
• Conventional VAEs have objectives misaligned with image fusion requirements
• Need to maintain fusion quality while significantly improving inference efficiency

Innovation

• Rectified Flow for one-step diffusion sampling
• Task-specific VAE with latent space fusion
• Two-stage training strategy for fusion optimization