Generative Adversarial Networks for single image with high quality image = Generative Adversarial Networks for single image with high quality image

다국어 초록 (Multilingual Abstract)

The SinGAN is one of generative adversarial networks that can be trained on a single nature image. It has poor ability to learn more global features from nature image, and losses much local detail information when it generates arbitrary size image sample. To solve the problem, a non-linear function is firstly proposed to control downsampling ratio that is ratio between the size of current image and the size of next downsampled image, to increase the ratio with increase of the number of downsampling. This makes the low-resolution images obtained by downsampling have higher proportion in all downsampled images. The low-resolution images usually contain much global information. Therefore, it can help the model to learn more global feature information from downsampled images. Secondly, the attention mechanism is introduced to the generative network to increase the weight of effective image information. This can make the network learn more local details. Besides, in order to make the output image more natural, the TVLoss function is introduced to the loss function of SinGAN, to reduce the difference between adjacent pixels and smear phenomenon for the output image. A large number of experimental results show that our proposed model has better performance than other methods in generating random samples with fixed size and arbitrary size, image harmonization and editing.

참고문헌 (Reference)

1 Arjovsky M, "Wasserstein generative adversarial networks" 70 : 214-223, 2017

2 Radford A, "Unsupervised representation learning with deep convolutional generative adversarial networks"

3 Li Y, "The Denoising of Desert Seismic Data Based on Cycle-GAN With Unpaired Data Training" 18 : 2016-2020, 2020

4 Jetchev N, "Texture synthesis with spatial generative adversarial networks"

5 Shaham T R, "Singan : Learning a generative model from a single natural image" 4569-4579, 2019

6 C. G. Turhan, "Recent Trends in Deep Generative Models : a Review" 574-579, 2018

7 Li C, "Precomputed Real-Time Texture Synthesis with Markovian Generative Adversarial Networks" 9907 : 702-716, 2016

8 Tilon S, "Post-Disaster Building Damage Detection from Earth Observation Imagery Using Unsupervised and Transferable Anomaly Detecting Generative Adversarial Networks" 12 : 4193-, 2020

9 Ledig C, "Photo-realistic single image super-resolution using a generative adversarial network" 105-114, 2017

10 Zhou Y, "Non-stationary texture synthesis by adversarial expansion" 37 (37): 1-13, 2018

11 Mao X, "Least Squares Generative Adversarial Networks" 2813-2821, 2017

12 Bergmann U, "Learning texture manifolds with the Periodic Spatial GAN" 70 : 469-477, 2017

13 Shocher A, "InGAN: Capturing and Remapping the “DNA” of a Natural Image"

14 Gulrajani I, "Improved training of wasserstein GANs" 5769-5779, 2017

15 Hinz T, "Improved Techniques for Training Single-Image GANs" 1300-1309, 2021

16 Yi X, "Generative adversarial network in medical imaging: A review" 58 : 2019

17 Goodfellow I J, "Generative adversarial nets" 27 : 2672-2680, 2014

18 Dong Y, "FD-GAN : Generative Adversarial Networks with Fusion-Discriminator for Single Image Dehazing" 10729-10736, 2020

19 Bin Wu, "Detecting Malicious Social Robots with Generative Adversarial Networks" 한국인터넷정보학회 13 (13): 5594-5615, 2019

20 Luan F, "Deep painterly harmonization" 37 (37): 95-106, 2018

21 Denton E, "Deep generative image models using a Laplacian pyramid of adversarial networks" 1 : 1486-1494, 2015

22 Mirza M, "Conditional generative adversarial nets"

23 Xin R, "Complex network classification with convolutional neural network" 25 (25): 447-457, 2020

24 Karras T, "A Style-Based Generator Architecture for Generative Adversarial Networks" 4396-4405, 2019

25 G. Bombuwala, "A Review of Generative Image Modeling Techniques" 1-5, 2019