Video Diffusion Models Enable Single-Image Refocusing

CSProfKGD · x · 2026-08-13

Adobe Research and partners introduced a novel method for post-capture image refocusing using Video Diffusion Models.

Core Approach: Generates a perceptually accurate focal stack (represented as a video sequence) from a single defocused image. It leverages the powerful priors of video diffusion models that naturally understand focus transitions.
Performance: Consistently outperforms existing approaches in perceptual quality and robustness across challenging real-world scenarios.
Open Resources: Released a large-scale focal stack dataset acquired under diverse smartphone conditions to support future research.

The paper will be presented at SIGGRAPH Asia 2025.

More from Multimodal