Densely Connected Pyramid Dehazing Network

By He Zhang, Vishal M. Patel. 2018

1,097 citations recorded by the source corpus.

We propose a new end-to-end single image dehazing method, called Densely Connected Pyramid Dehazing Network (DCPDN), which can jointly learn the transmission map, atmospheric light and dehazing all together. The end-to-end learning is achieved by directly embedding the atmospheric scattering model into the network, thereby ensuring that the proposed method strictly follows the physics-driven scattering model for dehazing. Inspired by the dense network that can maximize the information flow along features from different levels, we propose a new edge-preserving densely connected encoder-decoder structure with multi-level pyramid pooling module for estimating the transmission map. This network is optimized using a newly introduced edge-preserving loss function. To further incorporate the mutual structural information between the estimated transmission map and the dehazed result, we propose a joint-discriminator based on generative adversarial network framework to decide whether the corresponding dehazed image and the estimated transmission map are real or fake. An ablation study is conducted to demonstrate the effectiveness of each module evaluated at both estimated transmission map and dehazed result. Extensive experiments demonstrate that the proposed method achieves significant improvements over the state-of-the-art methods. Code and dataset is made available at: https://github.com/hezhangsprinter/DCPDN

Subfields: Computer Vision and Pattern Recognition

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

AttnGAN: Fine-Grained Text to Image Generation with Attentional Generative Adversarial Networks (2017, background): “Compared with other deep generative models, Generative Adversarial Networks (GANs) [6] have shown great performance for generating sharper samples [17, 3, 23, 13, 10, 35, 24, 34, 39, 40].”
Pre-Trained Image Processing Transformer (2020, background): “A various excellent works are also proposed for the other tasks, such as denoising [68, 32, 37, 45, 24], dehazing [6, 46, 85, 80], deraining [36, 78, 62, 29, 74, 47], and debluring [67, 53, 23, 10].”
A Generative Adversarial Approach for Zero-Shot Learning from Noisy Texts (2017, background): “Generative adversarial networks(GANs) [14] have shown promising performance on generating realistic images [47, 52, 53, 45, 30].”
Gated Fusion Network for Single Image Dehazing (2018, methodology): “To avoid hand-crafted priors, recent work [4, 14, 28, 41] automatically learns haze relevant features by convolutional neural networks (CNNs).”
DEA-Net: Single Image Dehazing Based on Detail-Enhanced Convolution and Content-Guided Attention (2023, methodology): “Inspired by their works [8], [32] and to take a further step, we design a Detail-Enhanced Convolution (DEConv) layer (see in Fig.”
Dual Residual Networks Leveraging the Potential of Paired Operations for Image Restoration (2019, methodology): “In subfigure (B), we show results for two real-world hazy images3 produced by two state-of-the-art methods, GFN [33] and DCPDN [51], and by ours.”
You Only Look Yourself: Unsupervised and Untrained Single Image Dehazing Neural Network (2020, methodology): “Evaluation Metrics: Like (Irani, 2019; Li et al., 2017; Qu et al., 2019; Ren et al., 2016; Zhang and Patel, 2018), two popular metrics are used in the quantitative comparisons, i.”
Translating and Segmenting Multimodal Medical Volumes with Cycle- and Shape-Consistency Generative Adversarial Network (2018, methodology): “In computer vision, recent image-to-image translation is formulated as a pixel-to-pixel mapping using encoderdecoder CNNs [16, 21, 43, 18, 44, 21, 38, 9].”
Vision Transformers for Single Image Dehazing (2022, methodology): “After that, DCPDN [10] uses two sub-networks to estimate t and A respectively, while GFN [12] estimates the fusion coefficient maps for the three predefined image operations.”
Domain Adaptation for Image Dehazing (2020, methodology): “Zhang and Patel [35] propose a densely connected pyramid network to estimate the transmission maps.”
Advancing Image Understanding in Poor Visibility Environments: A Collective Benchmark Study (2019, methodology): “Furthermore, as reported in [149] and Table VI, DCPDN has the best SSIM scores while MSCNN has the worst visual”
Gated Context Aggregation Network for Image Dehazing and Deraining (2018, methodology): “[42] design two different sub-networks for the prediction of the transmission map and the atmospheric light by following the physical model deﬁned in Equation (1).”
Real-World Underwater Enhancement: Challenges, Benchmarks, and Solutions Under Natural Light (2019, background): “Alternatively, the feature representation power of CNNs can also significantly improve the accuracy for depth or transmission estimation in complex scenes and uncontrolled outdoor environments [43].”
FFA-Net: Feature Fusion Attention Network for Single Image Dehazing (2019, methodology): “…estimate the haze effect, including the pioneering work of DehazeNet(Cai et al. 2016), the multi-scale CNN(MSCNN)(Ren et al. 2016), the residual learning technique(He et al. 2016), the quad-tree CNN(Kim, Ha, and Kwon 2018), and the densely connected pyramid dehazing network(Zhang and Patel 2018).”
Single Image Dehazing via Multi-scale Convolutional Neural Networks with Holistic Edges (2019, result): “Figure 14b–d show the results of DehazeNet (Cai et al. 2016), AOD-Net (Li et al. 2017), and DCPDN (Zhang and Patel 2018), respectively.”
Depth-Attentional Features for Single-Image Rain Removal (2019, methodology): “Moreover, since our network is also able to remove fog in the rain images, we further compared our method with two state-of-the-art haze removal methods, i.e., DCPDN [49] and AOD-Net [27]; they are both deep-learning-based methods, so we re-trained their models for rain removal.”
Face De-Spoofing: Anti-Spoofing via Noise Modeling (2018, methodology): “For de-mosaicing and super-resolution, many models, such as ResNet in [14,15] and joint models in [16,17,35], are learnt from the given pairs of low-quality input and high-quality ground truth.”
Multi-Scale Boosted Dehazing Network With Dense Feature Fusion (2020, methodology): “Method DCPDN [18] GDN [10] DuRN [11] MSBDN-DFF-S MSBDN-DFF Param 67M 1M 9M 4.”
Contrastive Learning for Compact Single Image Dehazing (2021, methodology): “Single image dehazing aims to generate the haze-free images from the hazy observation images, which can be categorized into prior-based methods [45, 17, 53, 4] and learning-based methods [5, 25, 51, 35, 23].”