Conditional GANs for Multi-Illuminant Color Constancy: Revolution or yet Another Approach?

Sidorov, Oleksii

doi:10.1109/cvprw.2019.00225

Cited by 41 publications

(37 citation statements)

References 58 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Wang et al (Wang et al 2018a) proposed a stacked conditional generative adversarial network (ST-CGAN) for image shadow removing. Sidorov (Sidorov 2019) proposed an end-to-end architecture named AngularGAN oriented specifically to the color constancy task, without estimating illumniation color or illumination color map. Wei et al (Wei et al 2019) proposed a two-stage generative adversarial network for shadow inpainting and removal with slice convolutions.…”

Section: Related Workmentioning

confidence: 99%

“…We compare our RIS-GAN with the state of-the-art methods including the two traditional methods, i.e., Guo (Guo, Dai, and Hoiem 2011) and Zhang (Zhang, Zhang, and Xiao 2015) and the recent learning-based methods, i.e., Deshad-owNet (Qu et al 2017), DSC (Hu et al 2018), ST-CGAN (Wang et al 2018a), and AngularGAN (Sidorov 2019). Note that shadow removal works on the pixel and recovers the value of the pixel, and therefore we add two frameworks, i.e., Global/Local-GAN (Iizuka, Simo-Serra, and Ishikawa 2017) for image inpainting and Pix2Pix-HD (Wang et al 2018b) for image translation, as another two shadow removal baselines for solid validation.…”

Section: Comparison With State-of-the-artsmentioning

confidence: 99%

“…A variety of existing works including traditional methods (Shor and Lischinski 2008;Xiao et al 2013a;, and learning-based methods (Gryka, Terry, and Brostow 2015;Wang et al 2018a;Le et al 2018) have been developed to solve this challenging problem. Different from traditional methods that highly rely on some prior knowledge (e.g., constant illumination and gradients) and often bring obvious artifacts on the shadow boundaries, learningbased methods especially recent deep learning methods like (Hu et al 2018) and (Sidorov 2019) have achieved some advances. However, the effectiveness of these methods highly depends on the training dataset and the designed network architectures.…”

Section: Introductionmentioning

confidence: 99%

“…In this paper, we propose a general framework RIS-GAN to explore both residual and illumination with Generative Adversarial Networks for shadow removal, unlike Sidorov's AngularGAN (Sidorov 2019) which just introduces an illumniation-based angular loss without estimating illumination color or illumination color map. As illustrated in Figure 1, our RIS-GAN consists of four generators in the encoder-decoder structure and three discriminators.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

RIS-GAN: Explore Residual and Illumination with Generative Adversarial Networks for Shadow Removal

Zhang

Long

Zhang

et al. 2020

AAAI

View full text Add to dashboard Cite

Residual images and illumination estimation have been proved very helpful in image enhancement. In this paper, we propose a general and novel framework RIS-GAN which explores residual and illumination with Generative Adversarial Networks for shadow removal. Combined with the coarse shadow-removal image, the estimated negative residual images and inverse illumination maps can be used to generate indirect shadow-removal images to refine the coarse shadow-removal result to the fine shadow-free image in a coarse-to-fine fashion. Three discriminators are designed to distinguish whether the predicted negative residual images, shadow-removal images, and the inverse illumination maps are real or fake jointly compared with the corresponding ground-truth information. To our best knowledge, we are the first one to explore residual and illumination for shadow removal. We evaluate our proposed method on two benchmark datasets, i.e., SRD and ISTD, and the extensive experiments demonstrate that our proposed method achieves the superior performance to state-of-the-arts, although we have no particular shadow-aware components designed in our generators.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Comparison With State-of-the-artsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

RIS-GAN: Explore Residual and Illumination with Generative Adversarial Networks for Shadow Removal

Zhang

Long

Zhang

et al. 2020

AAAI

View full text Add to dashboard Cite

show abstract

“…Instead of learning the shadow image directly, our network learns the shadow matting first and produces the high-quality shadow from the original shadow-free image and matting. Although these images are synthesized using neural network, they still have a similar distributions to the real natural scene compared with those from computer games (Sidorov 2019). In our shadow synthesis, we assume the scenes contain only cast shadows with the objects outside the scene.…”

Section: Introductionmentioning

confidence: 99%

Towards Ghost-Free Shadow Removal via Dual Hierarchical Aggregation Network and Shadow Matting GAN

Cun

Pun

Cheng

2020

AAAI

143

187

View full text Add to dashboard Cite

Shadow removal is an essential task for scene understanding. Many studies consider only matching the image contents, which often causes two types of ghosts: color in-consistencies in shadow regions or artifacts on shadow boundaries (as shown in Figure. 1). In this paper, we tackle these issues in two ways. First, to carefully learn the border artifacts-free image, we propose a novel network structure named the dual hierarchically aggregation network (DHAN). It contains a series of growth dilated convolutions as the backbone without any down-samplings, and we hierarchically aggregate multi-context features for attention and prediction, respectively. Second, we argue that training on a limited dataset restricts the textural understanding of the network, which leads to the shadow region color in-consistencies. Currently, the largest dataset contains 2k+ shadow/shadow-free image pairs. However, it has only 0.1k+ unique scenes since many samples share exactly the same background with different shadow positions. Thus, we design a shadow matting generative adversarial network (SMGAN) to synthesize realistic shadow mattings from a given shadow mask and shadow-free image. With the help of novel masks or scenes, we enhance the current datasets using synthesized shadow images. Experiments show that our DHAN can erase the shadows and produce high-quality ghost-free images. After training on the synthesized and real datasets, our network outperforms other state-of-the-art methods by a large margin. The code is available: http://github.com/vinthony/ghost-free-shadow-removal/

show abstract

Enhanced Shadow Removal for Surveillance Systems

Jishnu

Rajathilagam

2022

Innovative Data Communication Technologies and Application

View full text Add to dashboard Cite

Conditional GANs for Multi-Illuminant Color Constancy: Revolution or yet Another Approach?

Cited by 41 publications

References 58 publications

RIS-GAN: Explore Residual and Illumination with Generative Adversarial Networks for Shadow Removal

RIS-GAN: Explore Residual and Illumination with Generative Adversarial Networks for Shadow Removal

Towards Ghost-Free Shadow Removal via Dual Hierarchical Aggregation Network and Shadow Matting GAN

Enhanced Shadow Removal for Surveillance Systems

Contact Info

Product

Resources

About