Kindling the Darkness: A Practical Low-light Image Enhancer

By Yonghua Zhang, Jiawan Zhang, Xiaojie Guo. ACM Multimedia, 2019

1,723 citations recorded by the source corpus.

Images captured under low-light conditions often suffer from (partially) poor visibility. Besides unsatisfactory lightings, multiple types of degradations, such as noise and color distortion due to the limited quality of cameras, hide in the dark. In other words, solely turning up the brightness of dark regions will inevitably amplify hidden artifacts. This work builds a simple yet effective network for Kindling the Darkness (denoted as KinD), which, inspired by Retinex theory, decomposes images into two components. One component (illumination) is responsible for light adjustment, while the other (reflectance) for degradation removal. In such a way, the original space is decoupled into two smaller subspaces, expecting to be better regularized/learned. It is worth to note that our network is trained with paired images shot under different exposure conditions, instead of using any ground-truth reflectance and illumination information. Extensive experiments are conducted to demonstrate the efficacy of our design and its superiority over state-of-the-art alternatives. Our KinD is robust against severe visual defects, and user-friendly to arbitrarily adjust light levels. In addition, our model spends less than 50ms to process an image in VGA resolution on a 2080Ti GPU. All the above merits make our KinD attractive for practical use.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/28168e2c182e5456ad4712dae479dd44423b2ed6

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Low-Light Image Enhancement with Normalizing Flow (2021, methodology): “(Ying, Li, and Gao 2017) 15.95 0.6386 0.4573 DeepUPE (Wang et al. 2019a) 13.19 0.4902 0.4634 JED (Ren et al. 2018) 16.73 0.6817 0.3899 LIME (Guo, Li, and Ling 2016) 14.07 0.5274 0.4021 SICE (Cai, Gu, and Zhang 2018) 18.06 0.7094 0.5078 LLNet (Lore et al. 2017) 17.57 0.7388 0.4021 SRIE (Fu et al. 201”
SNR-Aware Low-light Image Enhancement (2022, methodology): “Methods Dong [9] LIME [15] MF [11] SRIE [12] BIMEF [55] DRD [45] RRM [28] SID [5] DeepUPE [40] KIND [61] DeepLPF [33] FIDE [48] PSNR 17.”
Learning Enriched Features for Fast Image Restoration and Enhancement (2022, methodology): “Notable works employ Retinex-inspired networks [4], [86], [87], [88], encoder-decoder networks [89], [90], [91], [92], [93], and GANs [94], [95], [96].”
URetinex-Net: Retinex-based Deep Unfolding Network for Low-light Image Enhancement (2022, methodology): “Furthermore, to verify the efficiency of our model in learning implicit prior from data, we compare with state-ofthe-art (SOTA) learning-based methods including RetinexNet [34], KinD [41], Zero-DCE [11], RUAS [24], AGLLNet [26] and KinD++ [40].”
Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement (2023, methodology): “These methods [54, 65, 66] usually suffer from a multi-stage training pipeline.”
Beyond Brightening Low-light Images (2021, methodology): “7 The network architectures of reflectance restoration used by our new KinD++ (a) and previous KinD (Zhang et al. 2019) (b)”
Learning a Simple Low-Light Image Enhancer from Paired Low-Light Instances (2023, methodology): “Among these approaches, most solutions rely on low-light and normal-light image pairs [33, 38].”
Sparse Gradient Regularized Deep Retinex Network for Robust Low-Light Image Enhancement (2021, methodology): “In [17], deep networks are designed to perform the Retinex decomposition and enhancement guided with the reﬂectance consistency constraint.”
Benchmarking Low-Light Image Enhancement and Beyond (2021, background): “…Robust Retinex Model (Robust) (Li et al. 2018), Single Image Contrast Enhancer (SICE) (Cai et al. 2018), Weighted Approximated Histogram Equalization (WAHE) (Arici et al. 2009), Kindling the Darkness (KinD) (Zhang et al. 2019), Deep Underexposed Photo Enhancement (DeepUPE) (Wang et al. 2019).”
Learning to Enhance Low-Light Image via Zero-Reference Deep Curve Estimation (2021, background): “[28] built a network for kindling the darkness of an image, called KinD, which decomposes images into two components.”
Retinex-inspired Unrolling with Cooperative Prior Architecture Search for Low-light Image Enhancement (2020, methodology): “KinD and DRBN indeed removed noises but introduced some unknown artifacts.”
An Experiment-Based Review of Low-Light Image Enhancement Methods (2020, background): “built a simple yet effective network called the Kindling the Darkness (KinD) network [272], which is composed of a layer decomposition net, a reflectance restoration net, and an illumination adjustment net, and trained it on pairs of images captured under different exposure conditions.”
MAXIM: Multi-Axis MLP for Image Processing (2022, background): “Visual examples for image low-light enhancement on the LOL dataset [96] between Retinex [96], GLAD [90], KinD [119], EnlightenGAN [36], MIRNet [107], and MAXIM-2S.”
Ultra-High-Definition Low-Light Image Enhancement: A Benchmark and Transformer-Based Method (2022, methodology): “For benchmarking, we compare 16 representative LLIE methods, including seven traditional non-learning methods (BIMEF (Ying, Li, and Gao 2017), FEA (Dong et al. 2011), LIME (Guo, Li, and Ling 2016), MF (Fu et al. 2016a), NPE (Wang et al. 2013), SRIE (Fu et al. 2016b), MSRCR (Jobson, Rahman, and Woode”