Hai-Rong Wang
ABSTRACT
With the rapid development of VR technology in the past two years, VR has been applied to many fields such as communication, video, game and so on. But for VR wearers, their faces are largely obscured, which hinders access to complete facial information. Aiming at the above problems, the neural network generated by cyclic confrontation is used to realize the restoration of the masked face. The neural network can learn the mapping from the face image with VR glasses to that without VR, GAN is used as the generating model, in which the discriminator can judge whether the image is real enough to ensure that the generated image will not lead to deformity, the improved CycleGAN model can make the map learn the distribution transformation of the image, generate the non-occlusion image from the occluded image, learn the mapping of each other, and guarantee not over-fitting. At the same time, the CycleGAN model is implemented by Pytorch Algorithm, and the trained model is applied to the test data set of 500 different faces of celebA.
- awaq Generative adversarial nets[C]// International Conference on Neural Information Processing Systems. MIT Press, 2014:2672-2680.Google Scholar
- Zhu J Y, Park T, Isola P, Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks[J]. 2017:2242-2251.Google Scholar
- Heusel M, Ramsauer H, Unterthiner T, Gans trained by a two time-scale update rule converge to a local nash equilibrium[C]//Advances in Neural Information Processing Systems.2017: 6626-6637.Google Scholar
- Mirza M, Osindero S. Conditional generative adversarial nets[J]. arXiv preprint arXiv:1411.1784, 2014.Google Scholar
- Arjovsky M, Chintala S, Bottou L. Wasserstein GAN[J]. 2017.Google Scholar
- Radford A, Metz L, Chintala S. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks[J]. Computer Science, 2015.Google Scholar
- Guoan Yang, Junjie Yang, Zhengzhi Lu A convolutional neural network with sparse representation[J] Knowledge-Based Systems, 2020, 209.Google Scholar
- JOHNSON J, ALAHI A, LI F F. Perceptual Losses for Real-Time Style Transfer and Super-Resolution[M]// Computer Vision-ECCV 2016. Springer International Publishing, 2016, 694-711.Google Scholar
- Karras T, Laine S, Aila T. A style-based generator architecture for generative adversarial networks[C]//Proceedings of the IEEEGoogle Scholar
- Conference on Computer Vision and Pattern Recognition. 2019:4401-4410.Google Scholar
- Hertzmann A, Jacobs C E, Oliver N, Image analogies[C]// Conference on Computer Graphics and Interactive Techniques. ACM, 2001:327-340.Google Scholar
- Isola P, Zhu J Y, Zhou T, Image-to-Image Translation with Conditional Adversarial Networks[J]. 2016:5967-5976.Google Scholar
- Pathak D, Krahenbuhl P, Donahue J, Context Encoders: Feature Learning by Inpainting[J].2016:2536-2544.Google Scholar
- Zhou T, Krahenbuhl P, Aubry M, Learning dense correspondence via 3d-guided cycle consistency[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016: 117-126.Google Scholar
- Kingma D P, Ba J. Adam: A Method for Stochastic Optimization[J]. Computer Science, 2014.Google Scholar
- Kingma D P, Welling M. Auto-Encoding Variational Bayes[J]. 2013.Google Scholar
Recommendations
Lifelong CycleGAN for continual multi-task image restoration
Highlights- A novel lifelong CycleGAN (LCGAN) framework is proposed to continually learn multiple image restoration tasks.
AbstractRecent years have witnessed the great success of deep learning in the applications of image restoration. However, there are still some challenging problems. First, most deep learning methods rely heavily on paired training images which ...
Image Restoration Using Gaussian Scale Mixtures in Complex Curvelet Transform Domain
ICMTMA '10: Proceedings of the 2010 International Conference on Measuring Technology and Mechatronics Automation - Volume 02In this paper, a complex Curvelet transform is presented at first. The key innovation can be generalized as follows:2D and 1D complex wavelet transform instead à trous algorithm sub-band decomposition and ID wavelet transform respectively, and increase ...
BIDGAN: Blind Image Deblurring with Improved CycleGAN and Frequency Filtering
Intelligent Computing Theories and ApplicationAbstractThe details of images can be severely impaired when they are captured in motion . To overcome such a blind image deblurring problem, we propose a deblurring method based on improved CycleGAN. This method can remarkably improve the definition of ...
Comments