GAN의 수정된 생성자에 의한 다중 도메인 스타일 트랜스퍼
A Multi-domain Style Transfer by Modified Generator of GAN
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 논문은 콘텐츠 이미지에 스타일 이미지를 입혀 스타일이 적용된 이미지를 생성하고자 기존의 이미지 대 이미지 방법이 아닌 다중 도메인 스타일 트랜스퍼(style transfer) 방법을 적용한다. 도메인별로 데이터 분포에 대한 다양한 특성을 고려하고, 콘텐츠 데이터의 특징을 잘 보존하면서높은 품질의 이미지가 생성되도록 잠재 벡터(latent vector)와 가우시안 노이즈를 추가하는 새로운GAN의 생성자(generator) 아키텍처를 제안한다. 콘텐츠 이미지가 도메인별 스타일을 잘 학습할 수있도록 네트워크를 구성하고 사계절 이미지로 구성된 도메인에 적용하여 고해상도의 스타일 트랜스퍼 결과를 보여준다.
Abstract
In this paper, we propose a novel generator architecture for multi-domain style transfer method not an image to image translation, as a method of generating a styled image by transfering a style to the content image. A latent vector and Gaussian noises are added to the generator of GAN so that a high quality image is generated while considering the characteristics of various data distributions for each domain and preserving the features of the content data. With the generator architecture of the proposed GAN, networks are configured and presented so that the content image can learn the styles for each domain well, and it is applied to the domain composed of images of the four seasons to show the high resolution style transfer results.
참고문헌 (24)
- A. Almahairi, "Augmented cyclegan: Learning many-to-many mappings from unpaired data", Proceeding of the 35th International Conference on Machine, pp.195~204 (2018)
- X. Huang, "Multimodal unsupervised image-to-image translation", ~ (2018)
- Y. Choi, "StarGAN v2: Diverse Image Synthesis for Multiple Domains", ~ (2020)
- T. Karras, "A Style-Based Generator architecture for Generative Adeversarial Networks", ~ (2019)
- S. Na, "MISO: Mutual Information Loss with Stochastic Style Representaions for Multimodal Image-to-Image Translation", ~ (2019)
- J-Y Zhu, "Toward Multimodal Image-to-Image Translation", ~ (2017)
- P. Isola, "Image-to-Image Translation with Conditional Adversarial Networks", ~ (2017)
- J-Y. Zhu, "Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks", ~ (2018)
- Q. Mao, "Mode Seeking Generative Adversarial Networks for Divese Image Synthesis", ~ (2019)
- Y. Choi, "StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation", ~ (2018)
- D. P. Kingma, "Auto-Encoding Variaitonal Bayes", ~ (2014)
- A. Odena, "Conditional Image Synthesis With Auxiliary Classifier GANs", ~ (2017)
- I. J. Goodfellow, "Generative Adversarial Nets", ~ (2014)
- L. A. Gatys, "Image Style Transfer Using Convolutional Neural Networks", Proceeding of IEEE Conference on Computer Vision and Pattern Recognition, pp.2414~2423 (2016)
- E. Risser, "Stable and Controllable Neural Texture Synthesis and Style Transfer Using Histogram Losses", ~ (2017)
- F. Luan, "Deep Photo Style Transfer", ~ (2017)
- X. Huang, "Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization", ~ (2017)
- S. Loffe, "Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift", ~ (2015)
- A. Radford, "Unsupervised representation learning with deep convolutional generative adversarial network", ~ (2016)
- D. Ylyanov, "Texture networks: Feed-forward synthesis of textures and stylized images", ~ (2016)
- D. Ulyanov, "Improved texture networks: Maximizing quality and diversity in feed-forward stylization and texture synthsesis", ~ (2017)
- Y. Shen, "InterFaceGAN:Interpreting the Disentangled Face Representation Learned by GANs", ~ (2020)
- J. Chang, "Layered Conceptual Image Compression Via Deep Semantic Synthesis", 2019. In Proceeding of the IEEE Conference on Image Processing, pp.694~698 (2019)
- M. Heusel, "GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium", ~ (2018)