• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.3
• /
• pp.109-114
• /
• 2021

GAN(Generative Adversarial Network) is an image generation model, which is composed of a generator network and a discriminator network, and generates an image similar to a real image. Since the image generated by the GAN should be similar to the actual image, a loss function is used to minimize the loss error of the generated image. However, there is a problem that the loss function of GAN degrades the quality of the image by making the learning to generate the image unstable. To solve this problem, this paper analyzes GAN-related studies and proposes an edge GAN(eGAN) using edge detection. As a result of the experiment, the eGAN model has improved performance over the existing GAN model.

• The Journal of the Convergence on Culture Technology
• /
• v.6 no.4
• /
• pp.767-776
• /
• 2020

GcGAN is a deep learning model to translate styles between images under geometric consistency constraint. However, GcGAN has a disadvantage that it does not properly maintain detailed content of an image, since it preserves the content of the image through limited geometric transformation such as rotation or flip. Therefore, in this study, we propose a new image-to-image translation method, MSGcGAN(Multi-Scale GcGAN), which improves this disadvantage. MSGcGAN, an extended model of GcGAN, performs style translation between images in a direction to reduce semantic distortion of images and maintain detailed content by learning multi-scale images simultaneously and extracting scale-invariant features. The experimental results showed that MSGcGAN was better than GcGAN in both quantitative and qualitative aspects, and it translated the style more naturally while maintaining the overall content of the image.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.1
• /
• pp.77-84
• /
• 2021

Today, the restoration of cultural properties is done by applying the latest IT technology from relying on existing data and experts. However, there are cases where new data are released and the original restoration is incorrect. Also, sometimes it takes too long to restore. And there is a possibility that the results will be different than expected. Therefore, we aim to quickly restore cultural properties using DeepLearning. Recently, so the algorithm DcGAN made in GANs algorithm, and image creation, restoring sectors are constantly evolving. We try to find the optimal GAN algorithm for the restoration of cultural properties among various GAN algorithms. Because the GAN algorithm is used in various fields. In the field of restoring cultural properties, it will show that it can be applied in practice by obtaining meaningful results. As a result of experimenting with the DCGAN and Style GAN algorithms among the GAN algorithms, it was confirmed that the DCGAN algorithm generates a top image with a low resolution.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.437-450
• /
• 2022

Generative adversarial networks(GANs) have demonstrated remarkable success in image synthesis. However, since GANs show instability in the training stage on large datasets, it is difficult to apply to various application fields. A single image GAN is a field that generates various images by learning the internal distribution of a single image. In this paper, we investigate five Single Image GAN: SinGAN, ConSinGAN, InGAN, DeepSIM, and One-Shot GAN. We compare the performance of each model and analyze the pros and cons of a single image GAN.

• Journal of the Korea Society of Computer and Information
• /
• v.25 no.5
• /
• pp.37-46
• /
• 2020

Recently, studies of facial attribute editing have obtained realistic results using generative adversarial net (GAN) and encoder-decoder structure. Spatial attention GAN (SaGAN), one of the latest researches, is the method that can change only desired attribute in a face image by spatial attention mechanism. However, sometimes unnatural results are obtained due to insufficient information on face areas. In this paper, we propose an improved SaGAN (MSaGAN) using a guide mask for learning and applying multitask learning approach to improve the limitations of the existing methods. Through extensive experiments, we evaluated the results of the facial attribute editing in therms of the mask loss function and the neural network structure. It has been shown that the proposed method can efficiently produce more natural results compared to the previous methods.

• Korean Journal of Remote Sensing
• /
• v.34 no.6_3
• /
• pp.1415-1425
• /
• 2018

Kompsat-5 is the first Earth Observation Satellite which is equipped with an SAR in Korea. SAR images are generated by receiving signals reflected from an object by microwaves emitted from a SAR antenna. Because the wavelengths of microwaves are longer than the size of particles in the atmosphere, it can penetrate clouds and fog, and high-resolution images can be obtained without distinction between day and night. However, there is no color information in SAR images. To overcome these limitations of SAR images, colorization of SAR images using Cycle GAN, a deep learning model developed for domain translation, was conducted. Training of Cycle GAN is unstable due to the unsupervised learning based on unpaired dataset. Therefore, we proposed MS Cycle GAN applying multi-scale discriminator to solve the training instability of Cycle GAN and to improve the performance of colorization in this paper. To compare colorization performance of MS Cycle GAN and Cycle GAN, generated images by both models were compared qualitatively and quantitatively. Training Cycle GAN with multi-scale discriminator shows the losses of generators and discriminators are significantly reduced compared to the conventional Cycle GAN, and we identified that generated images by MS Cycle GAN are well-matched with the characteristics of regions such as leaves, rivers, and land.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.11a
• /
• pp.41-43
• /
• 2019

Image-to-image 변환에서 인상적인 성능을 보이는 StarGAN 은 모델의 성능에 중요한 영향을 끼치는 adversarial weight, classification weight, reconstruction weight 라는 세가지 하이퍼파라미터의 결정을 전제로 하고 있다. 본 연구에서는 이 중 conditional GAN loss 인 adversarial loss 와 classification loss 를 대치할 수 있는 attribute loss를 제안함으로써, adversarial weight와 classification weight 를 최적화하는 데 걸리는 시간을 attribute weight 의 최적화에 걸리는 시간으로 대체하여 하이퍼파라미터 탐색에 걸리는 시간을 획기적으로 줄일 수 있게 하였다. 제안하는 attribute loss 는 각 특징당 GAN 을 만들 때 각 GAN 의 loss 의 합으로, 이 GAN 들은 hidden layer 를 공유하기 때문에 연산량의 증가를 거의 가져오지 않는다. 또한 reconstruction loss 를 단순화시켜 연산량을 줄인 simplified content loss 를 제안한다. StarGAN 의 reconstruction loss 는 generator 를 2 번 통과하지만 simplified content loss 는 1 번만 통과하기 때문에 연산량이 줄어든다. 또한 이미지 Framing 을 통해 배경의 왜곡을 방지하고, 양방향 성장을 통해 학습 속도를 향상시킨 아키텍쳐를 제안한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2021.11a
• /
• pp.930-932
• /
• 2021

영상처리에서 데이터 증강(Data augmentation)은 단순히 사진을 편집하여 사진의 개수를 증강하는 것이다. 단순 데이터 증강은 동물의 반점이나 다양한 색깔을 반영하지 못하는 한계가 있다. 본 논문에서는 GAN을 통한 데이터 증강 기법을 제안한다. 제안하는 방법은 CycleGAN을 사용하여 GAN 이미지를 생성한 뒤, 데이터 증강을 거쳐 동물의 종 분류 정확도를 측정한다. 정확도 비교를 위해 일반 사진으로만 구성한 집단과 GAN 사진을 추가한 두 집단으로 나누었다. ResNet50을 사용하여 종 분류 정확도를 측정한다.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.4
• /
• pp.292-302
• /
• 2020

Machine learning algorithms have made immense contributions in various fields including sonar and radar applications. Recently developed Cycle-Consistency Generative Adversarial Network (CycleGAN), a variant of GAN has been successfully used for unpaired image-to-image translation. We present a modified CycleGAN for translation of underwater ship engine sounds with high perceptual quality. The proposed network is composed of an improved generator model trained to translate underwater audio from one vessel type to other, an improved discriminator to identify the data as real or fake and a modified cycle-consistency loss function. The quantitative and qualitative analysis of the proposed CycleGAN are performed on publicly available underwater dataset ShipsEar by evaluating and comparing Mel-cepstral distortion, pitch contour matching, nearest neighbor comparison and mean opinion score with existing algorithms. The analysis results of the proposed network demonstrate the effectiveness of the proposed network.

• Journal of Information Technology Applications and Management
• /
• v.29 no.6
• /
• pp.81-93
• /
• 2022

As deep learning technology in natural language and visual processing has rapidly developed, collaborative filtering-based recommendation systems using deep learning technology are being actively introduced in the recommendation field. In this study, OCF-GAN, a hybrid collaborative filtering model using GAN, was proposed to solve the one-class and cold-start problems, and its usefulness was verified through performance evaluation. OCF-GAN based on conditional GAN consists of a generator that generates a pattern similar to the actual user preference pattern and a discriminator that tries to distinguish the actual preference pattern from the generated preference pattern. When the training is completed, user preference vectors are generated based on the actual distribution of preferred items. In addition, the cold-start problem was solved by using a hybrid collaborative filtering recommendation method that additionally utilizes user and item profiles. As a result of the performance evaluation, it was found that the performance of the OCF-GAN with additional information was superior in all indicators of the Top 5 and Top 20 recommendations compared to the existing GAN-based recommender. This phenomenon was more clearly revealed in experiments with cold-start users and items.