Single Image Dehazing via Conditional Generative Adversarial Network

By Runde Li, Jin-shan Pan, Zechao Li, Jinhui Tang. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018

405 citations recorded by the source corpus.

In this paper, we present an algorithm to directly restore a clear image from a hazy image. This problem is highly ill-posed and most existing algorithms often use hand-crafted features, e.g., dark channel, color disparity, maximum contrast, to estimate transmission maps and then atmospheric lights. In contrast, we solve this problem based on a conditional generative adversarial network (cGAN), where the clear image is estimated by an end-to-end trainable neural network. Different from the generative network in basic cGAN, we propose an encoder and decoder architecture so that it can generate better results. To generate realistic clear images, we further modify the basic cGAN formulation by introducing the VGG features and an L1-regularized gradient prior. We also synthesize a hazy dataset including indoor and outdoor scenes to train and evaluate the proposed algorithm. Extensive experimental results demonstrate that the proposed method performs favorably against the state-of-the-art methods on both synthetic dataset and real world hazy images.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/6a77a59257442da152f4e90729f483018659b7ef

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

Efficient and Explicit Modelling of Image Hierarchies for Image Restoration (2023, background): “Since the advent of deep learning, great efforts have been made to design CNNs for image restoration [28, 31, 42, 46, 72, 81, 82, 93, 96, 97].”
PSD: Principled Synthetic-to-Real Dehazing Guided by Physical Priors (2021, methodology): “In addition, some other methods [6, 14, 18, 21, 28, 29, 32] have been proposed to directly recover the clean images.”
All in One Bad Weather Removal Using Architectural Search (2020, background): “This is also true for bad weather restoration or image enhancement, such as dehazing [47, 3, 1, 57, 7, 13, 26, 43], removal of raindrop and dirt [20, 18, 59, 8, 53, 30, 50, 62, 36, 4], of moderate rain [44, 29], and of heavy rain [25, 54].”
Domain Adaptation for Image Dehazing (2020, methodology): “Also, some end-to-end methods [27, 16, 25] have been proposed to recover the clean image directly instead of estimating the transmission map and atmospheric light.”
Multi-Scale Boosted Dehazing Network With Dense Feature Fusion (2020, background): “Methods DCP [3] NLD [1] AODNet [6] MSCNN [13] MsPPN [19] DcGAN [8] GFN [14] GCANet [2] PFFNet [12] GDN [10] DuRN [11] Ours”
Gated Context Aggregation Network for Image Dehazing and Deraining (2018, methodology): “Loss Function In previous learning-based image dehazing methods [3, 31, 22, 24, 42, 44], the simple Mean Square Error loss is adopted.”
TransWeather: Transformer-based Restoration of Images Degraded by Adverse Weather Conditions (2021, background): “Adverse weather removal problems like deraining [16, 21,25,51,61,65,76], dehazing [1,2,10,22,42,69], desnowing [29,44,44,73] and rain drop removal [37,39,40,66] have been extensively explored in the literature.”
You Only Look Yourself: Unsupervised and Untrained Single Image Dehazing Neural Network (2020, background): “(2), J-Net takes a non-degenerate architecture by following (Li et al., 2018).”
Dual Residual Networks Leveraging the Potential of Paired Operations for Image Restoration (2019, methodology): “[24] proposed method to directly estimate J(x) without explicitly estimating t and A.”
Depth-Attentional Features for Single-Image Rain Removal (2019, background): “, snow [36], raindrops [34, 48], and haze [3, 27, 29, 35, 46, 49].”
EnlightenGAN: Deep Light Enhancement Without Paired Supervision (2019, methodology): “When applying GANs to image restoration and enhancement, most existing works use paired training data as well, such as super resolution [27], single image deblurring [28], deraining [29] and de-hazing [30].”