Range-Null Latent Prior-guided Consistency Model for Low Light Image Enhancement
Keywords: Low light enhancement, Consistency model
Abstract: Low light image enhancement (LLIE) is a challenging task, with most existing models often struggling to adapt to diverse dark environments due to insufficient training datasets. In this paper, we propose a novel unsupervised model called Range-null Latent Prior-guided Consistency Model (RLPCM), which integrates a latent consistency model (LCM) into low light enhancement using Retinex-based range-null space decomposition.RLPCM leverages an off-the-shelf LCM as a generative prior to improve both the latent consistency and realness of enhanced images. Meanwhile, fine-tuning a lighting decoder solely on normal-light images to ensure high fidelity in image space. A key contribution is a simple yet effective global illumination adjustment applied to the range-space component, along with a natural language guidance module to learn the null-space component. This allows for iterative generation to enhance both consistency and realness in just a few steps. Additionally, we present a new UAV low light dataset (UAV-LL) containing 300 image pairs from various UAV scenarios to support comprehensive evaluation. Extensive experiments demonstrate the superior adaptability and effectiveness of our framework across a wide range of low-light environments.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4060
Loading