MR. Illuminate: Zero-Shot Low-Light Image Enhancement with Diffusion Prior

The primary axes of interest in low-light image enhancement (LLIE) are color constancy-ensuring consistent outputs across inputs of the same scene under varying illumination and noise-and generalization across diverse datasets. Existing methods, whether supervised, unsupervised, or zero-shot, rely on auxiliary loss functions and empirically selected hyperparameters, which yield strong results on the datasets used for evaluation but often exhibit limited generalization. To overcome these constraints, we propose MR. Illuminate (pronounced "Mister Illuminate"), the first deep learning-based solution for LLIE that requires no optimization and no degradation assumption. "MR." emphasizes our Modulate-Refine design: global il-luminance and color are modulated via Adaptive Instance Normalization (AdaIN), while local structure and color are refined through self-attention features within a pre-trained diffusion model, taking a unique approach from prior methods. Extensive quantitative evaluations show that our approach surpasses SOTA methods on standard LLIE benchmarks, while qualitative results demonstrate improved color fidelity. Moreover, without any modification to our framework, our method achieves competitive results on the auto white balance (AWB) task, underscoring its strong generalization capability.

This CVPR paper is the Open Access version, provided by the Computer Vision Foundation.

Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore.

Figure 2. LLIE Method Taxonomy. LLIE approaches can be categorized by how and when learning occurs. A. Supervised methods, trained with paired datasets, learn an explicit image-to-image mapping through direct loss optimization. Unsupervised frameworks remove the need for paired supervision and instead learn domain-level correspondences between low-and normal-light distributions. Further details and limitations are discussed in Sec. C of the supplementary. B. Performs per-image optimization during test time, adjusting network weights according to a predefined loss. C. Leverages a frozen pre-trained prior while optimizing learnable components per input. However, both (B) and (C) are computationally expensive and sensitive to hyperparameter settings. D (Ours). In contrast to prior categories, our method leverages self-attention features derived directly from the input through a diffusion model to guide enhancement, and it yields consistent results for the same scene under varying illumination.