Residual diffusion enables faster and clearer image restoration
Residual Diffusion Implicit Models
Computer Vision and Pattern Recognition
Summary
Restoring damaged or low-quality images usually involves starting from random noise, which can lead to unwanted artifacts or 'hallucinations' in the final picture. The authors propose a new approach called residual diffusion implicit models (RDIMs) that focuses on modeling the exact difference between a good-quality and degraded image. This method allows the restoration to be done more accurately in just a few steps, often even in one, reducing errors and improving the quality of the results. RDIMs also provide flexible control over how much the process should focus on precision versus variety, making it useful for various image correction tasks.
What this means in practice
- •For image processing engineers: Improve automated tools for removing noise and enhancing resolution in photographs using fewer computational steps with better image fidelity.
- •For video game developers: Integrate faster image upscaling techniques that reduce artifacts and speed up rendering of high-quality textures from lower-resolution originals.
Authors
João Guerreiro, Pedro Tomás, Helena Aidos, Jacinto C. Nascimento
Abstract
Diffusion models achieve state-of-the-art results across multiple tasks. However, in inverse problems, standard initialization from pure Gaussian noise misaligns the generative process with real-world degradations. More recent methods such as diffusion bridges impose strict endpoint constraints and often require long reverse processes that are prone to hallucinations. Alternative consistency models provide noise-invariant, one-step mappings but lack inherent variance modeling and can degrade under severe corruption. Hence, residual diffusion implicit models (RDIMs) are proposed, constituting a generalized framework that explicitly models the residuals between high-quality (HQ) and low-quality (LQ) images, aligning the forward process with the actual degradation. A non-Markovian implicit reverse sampler is derived, which can skip intermediate timesteps, enabling accurate few-step or even single-step reconstruction, while mitigating the hallucinations inherent to long diffusion chains. RDIM also introduces a controllable variance mechanism that interpolates between deterministic and stochastic sampling, balancing fidelity and diversity. Furthermore, it enables the straightforward use of perceptual losses, when needed. Experiments on denoising and super-resolution benchmarks demonstrate that RDIMs consistently outperforms the state of the art, including bridge and consistency models, in terms of PSNR, SSIM, and LPIPS, reducing hallucinations while requiring only a few sampling steps (often just one). The results position RDIMs as an efficient solution for a broad range of image restoration tasks.