EnlightenGAN: Deep Light Enhancement Without Paired Supervision

  title={EnlightenGAN: Deep Light Enhancement Without Paired Supervision},
  author={Yi-fan Jiang and Xinyu Gong and Ding Liu and Yu Cheng and Chen Fang and Xiaohui Shen and Jianchao Yang and Pan Zhou and Zhangyang Wang},
  journal={IEEE Transactions on Image Processing},
Deep learning-based methods have achieved remarkable success in image restoration and enhancement, but are they still competitive when there is a lack of paired training data. [] Key Method Instead of supervising the learning using ground truth data, we propose to regularize the unpaired training using the information extracted from the input itself, and benchmark a series of innovations for the low-light image enhancement problem, including a global-local discriminator structure, a self-regularized…

Figures and Tables from this paper

Enhance Image as You Like with Unpaired Learning

A lightweight one-path conditional generative adversarial network (cGAN) is proposed to learn a one-to-many relation from low-light to normal-light image space, given only sets of low- and normal- light training images without any correspondence.

Contextual Information Aided Generative Adversarial Network for Low-Light Image Enhancement

Qualitative and quantitative experimental results demonstrate that the proposed improved generative adversarial network achieves relatively good results and has good generalization capacity compared to other state-of-the-art low-light enhancement algorithms.

Generative adversarial network for low-light image enhancement

Extensive experimental results show that the proposed end-to-end image enhancement method is capable of dealing with extremely low-light scenes and the realistic feature generator outperforms several state-of-the-art methods in a number of qualitative and quantitative evaluation tests.

Low Light Image Enhancement With Small Training Dataset Using EnlightenGAN

This study proposes EnlightenGAN in a smaller dataset environment to reduce training and processing time but with optimal results, and finds that the optimal minimum dataset number is above 50 training images.

From Fidelity to Perceptual Quality: A Semi-Supervised Approach for Low-Light Image Enhancement

A deep recursive band network (DRBN) is proposed to recover a linear band representation of an enhanced normal-light image with paired low/normal-light images, and then obtain an improved one by recomposing the given bands via another learnable linear transformation based on a perceptual quality-driven adversarial learning with unpaired data.

RetinexGAN:Unsupervised Low-Light Enhancement With Two-Layer Convolutional Decomposition Networks

Both quantitative and visual results show that RetinexGAN is largely superior to state-of-the-art methods.

A Two-Stage Unsupervised Approach for Low Light Image Enhancement

This paper proposes a two-stage unsupervised method that decomposes the low light image enhancement into a pre-enhancement and a post-refinement problem and shows that the method can significantly improve feature points matching and simultaneous localization and mapping in low light conditions.

Invertible Network for Unpaired Low-light Image Enhancement

This work proposes to leverage the invertible network to enhance low-light image in forward process and degrade the normal-light one inversely with unpaired learning, and designs various loss functions to ensure the stability of training and preserve more image details.

Low-Light Image Enhancement Based on Generative Adversarial Network

A low-light image enhancement technique (LIMET) with a fine-tuned conditional generative adversarial network is presented, which employs two discriminators to acquire a semantic meaning that imposes the obtained results to be realistic and natural.

Cycle-Interactive Generative Adversarial Network for Robust Unsupervised Low-Light Enhancement

A novel Cycle-Interactive Generative Adversarial Network for unsupervised low-light image enhancement, capable of not only better transferring illumination distributions between low/normal-light images but also manipulating detailed signals between two domains, e.g. suppressing/synthesizing realistic noise in the cyclic enhancement/degradation process is proposed.



Single Image Dehazing via Conditional Generative Adversarial Network

This paper proposes an algorithm to directly restore a clear image from a hazy image based on a conditional generative adversarial network (cGAN), where the clear image is estimated by an end-to-end trainable neural network.

Deep Photo Enhancer: Unpaired Learning for Image Enhancement from Photographs with GANs

An unpaired learning method for image enhancement based on the framework of two-way generative adversarial networks (GANs) with several improvements that significantly improve the stability of GAN training for this application.

LLNet: A deep autoencoder approach to natural low-light image enhancement

Unsupervised Class-Specific Deblurring

An end-to-end deblurring network designed specifically for a class of data that learns a strong prior on the clean image domain using adversarial loss and maps the blurred image to its clean equivalent and imposes a scale-space gradient error with an additional gradient module.

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

SRGAN, a generative adversarial network (GAN) for image super-resolution (SR), is presented, to its knowledge, the first framework capable of inferring photo-realistic natural images for 4x upscaling factors and a perceptual loss function which consists of an adversarial loss and a content loss.

Deep Retinex Decomposition for Low-Light Enhancement

Extensive experiments demonstrate that the proposed deep Retinex-Net learned on this LOw-Light dataset not only achieves visually pleasing quality for low-light enhancement but also provides a good representation of image decomposition.

Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training

This work proposes Disentangled Dehazing Network, an end-to-end model that generates realistic haze-free images using only unpaired supervision by introducing a physical-model based disentanglement and reconstruction mechanism.

Unsupervised Image Super-Resolution Using Cycle-in-Cycle Generative Adversarial Networks

This work proposes a Cycle-in-Cycle network structure with generative adversarial networks (GAN) as the basic component to tackle the single image super-resolution problem in a more general case that the low-/high-resolution pairs and the down-sampling process are unavailable.

Learning to See in the Dark

A pipeline for processing low-light images is developed, based on end-to-end training of a fully-convolutional network that operates directly on raw sensor data and replaces much of the traditional image processing pipeline, which tends to perform poorly on such data.

When Image Denoising Meets High-Level Vision Tasks: A Deep Learning Approach

This paper proposes a convolutional neural network for image denoising which achieves the state-of-the-art performance and proposes a deep neural network solution that cascades two modules for image Denoising and various high-level tasks, respectively, and uses the joint loss for updating only the denoise network via back-propagation.