Zero-Shot Image Dehazing

Li, Boyun; Gou, Yuanbiao; Liu, Zitao; Zhu, Hongyuan; Zhou, Joey Tianyi; Peng, Xi

doi:10.1109/tip.2020.3016134

Cited by 132 publications

(40 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We fuse the detail feature map into the coarse dehazed image from the dehazing backbone network and finally output a high-quality dehazed image. Since the encoder-decoder structure has been proved to achieve good results in image reconstruction fields [23], [21], many methods [24], [25] adopt the encoder-decoder structure as the estimation network for transmission map. Therefore, we use the encoder-decoder structure with residual connections as the dehazing network.…”

Section: Methodsmentioning

confidence: 99%

“…These methods have largely overcome the shortcomings of prior-based methods. Since the encoder-decoder structure has been proved to achieve good results in image reconstruction fields [23], [21], many methods [24], [25] adopt the encoder-decoder structure as the estimation network for the transmission map. The skip connections are usually established between the encoder and decoder in order to enable the decoder to take full advantage of the information in the encoder network.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Single Image Dehazing with An Independent Detail-Recovery Network

Li¹,

Cheng²,

Sun³

et al. 2021

Preprint

View full text Add to dashboard Cite

Single image dehazing is a prerequisite which affects the performance of many computer vision tasks and has attracted increasing attention in recent years. However, most existing dehazing methods emphasize more on haze removal but less on the detail recovery of the dehazed images. In this paper, we propose a single image dehazing method with an independent Detail Recovery Network (DRN), which considers capturing the details from the input image over a separate network and then integrates them into a coarse dehazed image. The overall network consists of two independent networks, named DRN and the dehazing network respectively. Specifically, the DRN aims to recover the dehazed image details through local and global branches respectively. The local branch can obtain local detail information through the convolution layer and the global branch can capture more global information by the Smooth Dilated Convolution (SDC). The detail feature map is fused into the coarse dehazed image to obtain the dehazed image with rich image details. Besides, we integrate the DRN, the physical-model-based dehazing network and the reconstruction loss into an end-to-end joint learning framework. Extensive experiments on the public image dehazing datasets (RESIDE-Indoor, RESIDE-Outdoor and the TrainA-TestA) illustrate the effectiveness of the modules in the proposed method and show that our method outperforms the state-of-theart dehazing methods both quantitatively and qualitatively. The code is released in https://github.com/YanLi-LY/Dehazing-DRN.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Single Image Dehazing with An Independent Detail-Recovery Network

Li¹,

Cheng²,

Sun³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The DPATN demonstrated impressive dehazing performance, even in distant regions; however, it might suffer from post-dehazing artifacts. Currently, Li et al [105] exploited zero-shot learning to devise a training-free unsupervised dehazing network. They utilized three encoder-decoder-based submodules, known as J-Net, T-Net, and A-Net, corresponding to three unknowns in the simplified Koschmieder model.…”

Section: Deep Learningmentioning

confidence: 99%

Visibility Restoration: A Systematic Review and Meta-Analysis

Ngo

Lee

Ngo

et al. 2021

Sensors

View full text Add to dashboard Cite

Image acquisition is a complex process that is affected by a wide variety of internal and environmental factors. Hence, visibility restoration is crucial for many high-level applications in photography and computer vision. This paper provides a systematic review and meta-analysis of visibility restoration algorithms with a focus on those that are pertinent to poor weather conditions. This paper starts with an introduction to optical image formation and then provides a comprehensive description of existing algorithms as well as a comparative evaluation. Subsequently, there is a thorough discussion on current difficulties that are worthy of a scientific effort. Moreover, this paper proposes a general framework for visibility restoration in hazy weather conditions while using haze-relevant features and maximum likelihood estimates. Finally, a discussion on the findings and future developments concludes this paper.

show abstract

“…On the other hand, Chaitanya and Mukherjee [ 29 ] and Sun et al [ 30 ] exploited a cycle-consistent adversarial network, widely referred to as CycleGAN, to facilitate the use of unpaired datasets. Notably, Li et al [ 31 ] leveraged zero-shot learning, currently in its infancy, to fully relax a paired dataset requirement. They designed three networks that operate on the input image to estimate the scene radiance, transmittance, and global lightness, respectively.…”

Section: Literature Reviewmentioning

confidence: 99%

Automating a Dehazing System by Self-Calibrating on Haze Conditions

Ngo

Lee

et al. 2021

Sensors

View full text Add to dashboard Cite

Existing image dehazing algorithms typically rely on a two-stage procedure. The medium transmittance and lightness are estimated in the first stage, and the scene radiance is recovered in the second by applying the simplified Koschmieder model. However, this type of unconstrained dehazing is only applicable to hazy images, and leads to untoward artifacts in haze-free images. Moreover, no algorithm that can automatically detect the haze density and perform dehazing on an arbitrary image has been reported in the literature to date. Therefore, this paper presents an automated dehazing system capable of producing satisfactory results regardless of the presence of haze. In the proposed system, the input image simultaneously undergoes multiscale fusion-based dehazing and haze-density-estimating processes. A subsequent image blending step then judiciously combines the dehazed result with the original input based on the estimated haze density. Finally, tone remapping post-processes the blended result to satisfactorily restore the scene radiance quality. The self-calibration capability on haze conditions lies in using haze density estimate to jointly guide image blending and tone remapping processes. We performed extensive experiments to demonstrate the superiority of the proposed system over state-of-the-art benchmark methods.

show abstract

Zero-Shot Image Dehazing

Cited by 132 publications

References 30 publications

Single Image Dehazing with An Independent Detail-Recovery Network

Single Image Dehazing with An Independent Detail-Recovery Network

Visibility Restoration: A Systematic Review and Meta-Analysis

Automating a Dehazing System by Self-Calibrating on Haze Conditions

Contact Info

Product

Resources

About