Adversarial Edge-Aware Image Colorization With Semantic Segmentation

Kong, Guangqian; Tian, Huan; Duan, Xun; Long, Huiyun

doi:10.1109/access.2021.3056144

Cited by 17 publications

(7 citation statements)

References 31 publications

(49 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They train the network by minimizing the cross-entropy between per pixel color distributions and L2 loss on the chrominance channels. [Kong et al, 2021] propose to colorize a grayscale image by training a multitask network for colorization and semantic segmentation in an adversarial manner. They train a U-Net type network with a three term cost function: a color regression loss in terms of hue, saturation and lightness, the cross-entropy on the ground truth and generated semantic labels, and a GANs term.…”

Section: Deep Learning Methods For Image Colorizationmentioning

confidence: 99%

“…Histogram User Diverse Object Survey GANs prediction guided aware [Cheng et al, 2015] [Iizuka et al, 2016] [Vitoria et al, 2020] [Nazeri et al, 2018] [Cao et al, 2017] [Yoo et al, 2019] [Antic, 2019] [Larsson et al, 2016] [Zhang et al, 2016] [Mouzon et al, 2019] [He et al, 2018] [Deshpande et al, 2017] [Guadarrama et al, 2017] [Royer et al, 2017] [Kumar et al, 2021] [Su et al, 2020] [Pucci et al, 2021 [ Kong et al, 2021] [ Gu et al, 2019] (winner)…”

Section: Usingmentioning

confidence: 99%

“…These effects are independent from the color space or the loss. Some methods reduce this effect by introducing semantic information (e.g., [Vitoria et al, 2020]) or spatial localization (e.g., [Su et al, 2020]), while others achieve to reduce it by considering segmentation as an additional task (e.g., [Kong et al, 2021]). Note that with the VGG-based LPIPS, Lab color space provides more realistic result on the tennisman image.…”

Section: Qualitative Evaluationmentioning

confidence: 99%

See 2 more Smart Citations

Influence of Color Spaces for Deep Learning Image Colorization

Ballester¹,

Bugeau²,

Carrillo³

et al. 2022

Preprint

View full text Add to dashboard Cite

Colorization is a process that converts a grayscale image into a color one that looks as natural as possible. Over the years this task has received a lot of attention. Existing colorization methods rely on different color spaces: RGB, YUV, Lab, etc. In this chapter, we aim to study their influence on the results obtained by training a deep neural network, to answer the question: "Is it crucial to correctly choose the right color space in deep-learning based colorization?". First, we briefly summarize the literature and, in particular, deep learning-based methods. We then compare the results obtained with the same deep neural network architecture with RGB, YUV and Lab color spaces. Qualitative and quantitative analysis do not conclude similarly on which color space is better. We then show the importance of carefully designing the architecture and evaluation protocols depending on the types of images that are being processed and their specificities: strong/small contours, few/many objects, recent/archive images.

show abstract

Section: Deep Learning Methods For Image Colorizationmentioning

confidence: 99%

Section: Usingmentioning

confidence: 99%

Section: Qualitative Evaluationmentioning

confidence: 99%

See 1 more Smart Citation

Influence of Color Spaces for Deep Learning Image Colorization

Ballester¹,

Bugeau²,

Carrillo³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…[ Kong et al, 2021] proposes a multitask network in an adversarial manner that uses a MSE loss on hue, saturation and lightness channels to perform colorization and a Cross-Entropy loss to learn a semantic segmentation.…”

Section: Distribution-based Lossesmentioning

confidence: 99%

Analysis of Different Losses for Deep Learning Image Colorization

Ballester¹,

Bugeau²,

Carrillo³

et al. 2022

Preprint

View full text Add to dashboard Cite

Image colorization aims to add color information to a grayscale image in a realistic way. Recent methods mostly rely on deep learning strategies. While learning to automatically colorize an image, one can define well-suited objective functions related to the desired color output. Some of them are based on a specific type of error between the predicted image and ground truth one, while other losses rely on the comparison of perceptual properties. But, is the choice of the objective function that crucial, i.e., does it play an important role in the results? In this chapter, we aim to answer this question by analyzing the impact of the loss function on the estimated colorization results. To that goal, we review the different losses and evaluation metrics that are used in the literature. We then train a baseline network with several of the reviewed objective functions: classic L1 and L2 losses, as well as more complex combinations such as Wasserstein GAN and VGG-based LPIPS loss. Quantitative results show that the models trained with VGG-based LPIPS provide overall slightly better results for most evaluation metrics. Qualitative results exhibit more vivid colors when with Wasserstein GAN plus the L2 loss or again with the VGG-based LPIPS. Finally, the convenience of quantitative user studies is also discussed to overcome the difficulty of properly assessing on colorized images, notably for the case of old archive photographs where no ground truth is available.

show abstract

“…Effective use of modern information technology means, improving the quality of the work efficiency, to ensure the smooth running of the games, has become an important part of the modern comprehensive games. With the increasing functions and organizational workload of sports events, widely with the public and the news media, information technology also needs to fit modern sporting events of the practical need of the introduction of new design and modern information technology to improve the quality of work [ 13 , 14 ].…”

Section: Introductionmentioning

confidence: 99%

Design and Application of Major Sports Events Management Information System Based on Integration Algorithm

Wang

2022

Computational Intelligence and Neuroscience

View full text Add to dashboard Cite

With the rapid development of social economy and the change of ideology, people are increasingly enthusiastic about participating in sports events. How to make better use of the existing massive network data of sports events to solve the overload of information circulation of sports events has become a potential application problem to promote the development of sports information digitalization. Based on this problem orientation, this paper chooses to study the common characteristics of the current Internet data of sports events and the corresponding and applicable event recommendation technology and constructs feasible event recommendation model as the main direction of solution. The competition information system is the core technology system of the games and the central nervous system of the games. By summing up the previous high-level comprehensive games, the competition information system should serve the competition, command, media, and the public. The smooth construction and stable operation of the perfect design of the competition information system will reflect the organization level of the games and guarantee the success of the games. At the same time, it has certain guiding significance for the design of the competition information system of the large-scale comprehensive games in China in the future.

show abstract

Adversarial Edge-Aware Image Colorization With Semantic Segmentation

Cited by 17 publications

References 31 publications

Influence of Color Spaces for Deep Learning Image Colorization

Influence of Color Spaces for Deep Learning Image Colorization

Analysis of Different Losses for Deep Learning Image Colorization

Design and Application of Major Sports Events Management Information System Based on Integration Algorithm

Contact Info

Product

Resources

About