RealVDeblur: One-Step Diffusion for Generalizable Real-World Video Deblurring

2026-07-22

AI Summary

This work addresses the challenges of real-world video deblurring—namely, complex motion patterns, diverse degradation types, and the scarcity of authentic training data—by introducing a novel approach that integrates physically realistic blur synthesis with an efficient generative prior. Realistic blurred data is constructed using 3D Gaussian splatting and high-frame-rate videos, while a frame-level encoded video diffusion model, operating without VAE temporal compression, serves as the generative prior. The multi-step diffusion process is distilled into a single-step generator to enable efficient inference. A training-agnostic temporal window masking mechanism is innovatively introduced, allowing stable processing of long videos under constant memory constraints. By uniquely combining single-step diffusion with physically grounded synthetic data, the method significantly enhances out-of-domain generalization, achieving state-of-the-art perceptual quality, semantic fidelity, and temporal consistency on multiple real-world benchmarks, while also improving robustness in 3D reconstruction under severe motion blur.

Abstract

Real-world video deblurring remains challenging due to diverse motion patterns, complex degradations, and the scarcity of realistic training data, yet robust restoration is critical for downstream pipelines such as mobile imaging and 3D reconstruction. This work presents \textbf{RealVDeblur}, an efficient generative framework designed to improve in-the-wild robustness under diverse real capture conditions. First, a large-scale, physically grounded blur synthesis pipeline is constructed from scene-level 3D Gaussian Splatting (3DGS) assets and high-frame-rate videos, providing realistic training data covering both camera-induced and object-motion blur. Second, a video diffusion prior is leveraged for restoration; to better accommodate frame-dependent blur variations, temporal compression in the VAE is disabled and a frame-wise encoding scheme is adopted. For practical deployment on long videos, multi-step diffusion sampling is distilled into an efficient one-step generator, and a training-free Temporal Window Mask stabilizes inference beyond the training horizon with constant memory usage. Extensive experiments on diverse real-world benchmarks demonstrate strong perceptual quality, semantic fidelity, and temporal consistency on unseen videos, as well as improved robustness in downstream 3D reconstruction under severe motion blur. Project page: https://rbjin.github.io/RealVDeblur

Problem

• video deblurring
• real-world degradation
• motion blur
• training data scarcity
• temporal consistency

Innovation

• video deblurring
• diffusion model
• 3D Gaussian Splatting
• one-step generation
• temporal consistency