I-HAZE: A Dehazing Benchmark with Real Hazy and Haze-Free Indoor Images

Ancuti, Cosmin; Timofte, Radu; Vleeschouwer, Christophe De

doi:10.1007/978-3-030-01449-0_52

Cited by 243 publications

(141 citation statements)

References 47 publications

Supporting

Mentioning

119

Contrasting

Unclassified

Order By: Relevance

“…The first experiment is to compare our result with the state-of-the-art approaches on NYU-Depth [28] dataset. Then, we have investigated our performance on the NTIRE 2018 challenge on single image dehazing [4] datasets: I-HAZE [6] & O-HAZE [7]. In addition, we have emphasized differences between CycleGAN [37] and our proposed method, Cycle-Dehaze, via qualitative and quantitative results.…”

Section: Experiments and Resultsmentioning

confidence: 99%

“…We have focused on the NTIRE 2018 challenge on single image dehazing [4] datasets: I-HAZE [6] & O-HAZE [7] during the preparation of this work. We have analyzed effects of Laplacian pyramid and cyclic perceptual-loss, especially on I-HAZE [6] dataset. The challenge datasets are considerably higher resolution than other image dehazing datasets e.g.…”

Section: Results On I-haze and O-haze Datasetsmentioning

confidence: 99%

“…This shows that additional cyclic perceptual-consistency loss and Laplacian pyramid increase the performance of original CycleGAN [37] architecture. Moreover, PSNR and SSIM differences between the I-HAZE [6] and O-HAZE [7] datasets presents that outdoor scenes suffer from SSIM values because of the long shot of the captured images. On the other hand, they have higher PSNR values since the produced fog spreads the atmosphere and the captured images seem less hazed than the indoor scenes.…”

Section: Results On I-haze and O-haze Datasetsmentioning

confidence: 99%

“…Each image includes a Macbeth ColorChecker mostly used for color calibration with the real-world. The challenge [4] has two main datasets: I-HAZE [6] & O-HAZE [7], which have 25 indoor and 35 outdoor hazy images and their ground truth images, respectively. The captured images are in very high resolution.…”

Section: Nyu-depthmentioning

confidence: 99%

See 3 more Smart Citations

Cycle-Dehaze: Enhanced CycleGAN for Single Image Dehazing

Deniz

Genç

Ekenel

2018

2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

451

239

View full text Add to dashboard Cite

In this paper, we present an end-to-end network, called Cycle-Dehaze, for single image dehazing problem, which does not require pairs of hazy and corresponding ground truth images for training. That is, we train the network by feeding clean and hazy images in an unpaired manner. Moreover, the proposed approach does not rely on estimation of the atmospheric scattering model parameters. Our method enhances CycleGAN formulation by combining cycle-consistency and perceptual losses in order to improve the quality of textural information recovery and generate visually better haze-free images. Typically, deep learning models for dehazing take low resolution images as input and produce low resolution outputs. However, in the NTIRE 2018 challenge on single image dehazing, high resolution images were provided. Therefore, we apply bicubic downscaling. After obtaining low-resolution outputs from the network, we utilize the Laplacian pyramid to upscale the output images to the original resolution. We conduct experiments on NYU-Depth, I-HAZE, and O-HAZE datasets. Extensive experiments demonstrate that the proposed approach improves CycleGAN method both quantitatively and qualitatively.

show abstract

Section: Experiments and Resultsmentioning

confidence: 99%

Section: Results On I-haze and O-haze Datasetsmentioning

confidence: 99%

Section: Results On I-haze and O-haze Datasetsmentioning

confidence: 99%

Section: Nyu-depthmentioning

confidence: 99%

See 2 more Smart Citations

Cycle-Dehaze: Enhanced CycleGAN for Single Image Dehazing

Deniz

Genç

Ekenel

2018

2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

451

239

View full text Add to dashboard Cite

show abstract

“…The dataset was distributed by NTIRE 2018 Challenge on image dehazing [19]. Two novel subsets (I-HAZE [20] and O-HAZE [21]) with real haze and their ground-truth haze-free images were included. Hazy images were both captured in presence of real haze generated by professional haze machines.…”

Section: Ntire2018 Image Dehazing Datasetmentioning

confidence: 99%

Progressive Feature Fusion Network for Realistic Image Dehazing

Mei

Jiang

et al. 2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

0000−0001−8949−9597] , Aiwen Jiang 1[0000−0002−5979−7590] , Juncheng Li 2[0000−0001−7314−6754] , and Abstract. Single image dehazing is a challenging ill-posed restoration problem. Various prior-based and learning-based methods have been proposed. Most of them follow a classic atmospheric scattering model which is an elegant simplified physical model based on the assumption of singlescattering and homogeneous atmospheric medium. The formulation of haze in realistic environment is more complicated. In this paper, we propose to take its essential mechanism as "black box", and focus on learning an input-adaptive trainable end-to-end dehazing model. An U-Net like encoder-decoder deep network via progressive feature fusions has been proposed to directly learn highly nonlinear transformation function from observed hazy image to haze-free ground-truth. The proposed network is evaluated on two public image dehazing benchmarks. The experiments demonstrate that it can achieve superior performance when compared with popular state-of-the-art methods. With efficient GPU memory usage, it can satisfactorily recover ultra high definition hazed image up to 4K resolution, which is unaffordable by many deep learning based dehazing algorithms.

show abstract