One-Step Diffusion Model for Image Motion-Deblurring

2025-03-09

AI Summary

Existing single-image deblurring methods suffer from limitations in perceptual quality, generalization capability, and robustness to complex blur patterns. While diffusion models show promise, their multi-step sampling process severely hampers inference efficiency. To address this, we propose OSDD—the first one-step diffusion-based deblurring model—that compresses iterative denoising into a single inference step, achieving both high-fidelity reconstruction and significant speedup. Our key contributions are: (1) the first unified one-step diffusion deblurring framework; (2) an enhanced variational autoencoder (eVAE) that improves structural recovery; and (3) a dynamic dual adapter (DDA) that jointly optimizes perceptual quality and pixel-level fidelity. OSDD achieves state-of-the-art performance on both full-reference (e.g., PSNR, LPIPS) and no-reference metrics (e.g., NIQE, BRISQUE), while accelerating inference by over an order of magnitude compared to typical diffusion-based approaches. Code and pretrained models are publicly available.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Currently, methods for single-image deblurring based on CNNs and transformers have demonstrated promising performance. However, these methods often suffer from perceptual limitations, poor generalization ability, and struggle with heavy or complex blur. While diffusion-based methods can partially address these shortcomings, their multi-step denoising process limits their practical usage. In this paper, we conduct an in-depth exploration of diffusion models in deblurring and propose a one-step diffusion model for deblurring (OSDD), a novel framework that reduces the denoising process to a single step, significantly improving inference efficiency while maintaining high fidelity. To tackle fidelity loss in diffusion models, we introduce an enhanced variational autoencoder (eVAE), which improves structural restoration. Additionally, we construct a high-quality synthetic deblurring dataset to mitigate perceptual collapse and design a dynamic dual-adapter (DDA) to enhance perceptual quality while preserving fidelity. Extensive experiments demonstrate that our method achieves strong performance on both full and no-reference metrics. Our code and pre-trained model will be publicly available at https://github.com/xyLiu339/OSDD.

Problem

• Single-step diffusion model for efficient image deblurring
• Enhanced variational autoencoder improves structural restoration
• Dynamic dual-adapter enhances perceptual quality and fidelity

Innovation

• One-step diffusion model for efficient deblurring
• Enhanced variational autoencoder improves structural restoration
• Dynamic dual-adapter enhances perceptual quality