• International journal of advanced smart convergence
• /
• 제8권1호
• /
• pp.141-146
• /
• 2019

These days, there are many applications using neural networks as parts of their system. On the other hand, adversarial examples have become an important issue concerining the security of neural networks. A classifier in neural networks can be fooled and make it miss-classified by adversarial examples. There are many research to encounter adversarial examples by using denoising methods. Some of them using GAN (Generative Adversarial Network) in order to remove adversarial noise from input images. By producing an image from generator network that is close enough to the original clean image, the adversarial examples effects can be reduced. However, there is a chance when adversarial noise can survive the approximation process because it is not like a normal noise. In this chance, we propose a research that utilizes high-level representation in the classifier by combining GAN network with a trained U-Net network. This approach focuses on minimizing the loss function on high representation terms, in order to minimize the difference between the high representation level of the clean data and the approximated output of the noisy data in the training dataset. Furthermore, the generated output is checked whether it shows minimum error compared to true label or not. U-Net network is trained with true label to make sure the generated output gives minimum error in the end. At last, the remaining adversarial noise that still exist after low-level approximation can be removed with the U-Net, because of the minimization on high representation terms.

• 한국인터넷방송통신학회논문지
• /
• 제20권1호
• /
• pp.239-246
• /
• 2020

최근에 제안된 WGAN(Wasserstein generative adversarial network)의 등장으로 GAN(generative adversarial network)의 고질적인 문제인 까다롭고 불안정한 학습과정이 다소 개선되기는 하였으나 여전히 수렴이 안되거나 자연스럽지 못한 출력물을 생성하는 등의 경우가 발생한다. 이러한 문제를 해결하기 위하여 본 논문에서는 분별기가 실제 데이터 확률분포를 보다 정확히 추정할 수 있도록 표본화 과정을 개선하는 동시에 분별기 함수의 립쉬츠 연속조건을 안정적으로 유지시키기 위한 알고리즘을 제안한다. 다양한 실험을 통하여 제안 기법의 특성을 분석하고 성능을 확인한다.

• International Journal of Internet, Broadcasting and Communication
• /
• 제11권4호
• /
• pp.31-36
• /
• 2019

Unsupervised neural networks have not caught enough attention until Generative Adversarial Network (GAN) was proposed. By using both the generator and discriminator networks, GAN can extract the main characteristic of the original dataset and produce new data with similarlatent statistics. However, researchers understand fully that training GAN is not easy because of its unstable condition. The discriminator usually performs too good when helping the generator to learn statistics of the training datasets. Thus, the generated data is not compelling. Various research have focused on how to improve the stability and classification accuracy of GAN. However, few studies delve into how to improve the training efficiency and to save training time. In this paper, we propose a novel optimizer, named FAST-ADAM, which integrates the Lookahead to ADAM optimizer to train the generator of a semi-supervised generative adversarial network (SSGAN). We experiment to assess the feasibility and performance of our optimizer using Canadian Institute For Advanced Research - 10 (CIFAR-10) benchmark dataset. From the experiment results, we show that FAST-ADAM can help the generator to reach convergence faster than the original ADAM while maintaining comparable training accuracy results.

• 스마트미디어저널
• /
• 제13권3호
• /
• pp.27-35
• /
• 2024

본 연구는 GAN(Generative Adversarial Network)을 기반으로 한 Semi Supervised Learning을 활용하여 이미지 생성과 ResNet50을 이용한 이미지 분류를 결합하는 방법에 대해 다루고 있다. 이를 통해 새로운 접근법을 제시하여 이미지 생성과 분류를 통합함으로써 더 정확하고 다양한 결과를 얻을 수 있도록 하였다. 생성자와 판별자를 학습시켜 생성된 이미지와 실제 이미지를 구별하고, ResNet50을 활용하여 이미지 분류를 수행한다. 실험 결과에서는 생성된 이미지의 품질이 epoch에 따라 변화함을 확인할 수 있었으며, 이를 통해 산업재해 예측 정확성을 향상하고자 한다. 또한, GAN과 ResNet50의 결합을 통해 이미지 생성의 품질을 향상시키고 이미지 분류의 정확도를 높이는 효율적인 방법을 제시하고자 한다.

• International Journal of Computer Science & Network Security
• /
• 제24권2호
• /
• pp.136-142
• /
• 2024

Haze is a very common phenomenon that degrades or reduces the visibility. It causes various problems where high quality images are required such as traffic and security monitoring. So haze removal from images receives great attention for clear vision. Due to its huge impact, significant advances have been achieved but the task yet remains a challenging one. Recently, different types of deep generative adversarial networks (GAN) are applied to suppress the noise and improve the dehazing performance. But it is unclear how these algorithms would perform on hazy images acquired "in the wild" and how we could gauge the progress in the field. This paper aims to bridge this gap. We present a comprehensive study and experimental evaluation on diverse GAN models in single image dehazing through benchmark datasets.

• 한국BIM학회 논문집
• /
• 제9권1호
• /
• pp.42-51
• /
• 2019

Maintenance of aging structures has attracted societal attention. Maintenance of the aging structure can be efficiently performed with a digital twin. In order to maintain the structure based on the digital twin, it is required to accurately detect the damage of the structure. Meanwhile, deep learning-based damage detection approaches have shown good performance for detecting damage of structures. However, in order to develop such deep learning-based damage detection approaches, it is necessary to use a large number of data before and after damage, but there is a problem that the amount of data before and after the damage is unbalanced in reality. In order to solve this problem, this study proposed a method based on Generative adversarial network, one of Generative Model, for generating acceleration data usually used for damage detection approaches. As results, it is confirmed that the acceleration data generated by the GAN has a very similar pattern to the acceleration generated by the simulation with structural analysis software. These results show that not only the pattern of the macroscopic data but also the frequency domain of the acceleration data can be reproduced. Therefore, these findings show that the GAN model can analyze complex acceleration data on its own, and it is thought that this data can help training of the deep learning-based damage detection approaches.

• 디지털콘텐츠학회 논문지
• /
• 제19권3호
• /
• pp.435-443
• /
• 2018

최근 컴퓨팅 파워의 폭발적인 발전으로 컴퓨팅의 한계 라는 장벽이 사라지면서 딥러닝 이라는 이름 하에 순환 신경망(RNN), 합성곱 신경망(CNN) 등 다양한 모델들이 제안되어 컴퓨터 비젼(Computer Vision)의 수많은 난제들을 풀어나가고 있다. 2014년 발표된 대립쌍 모델(Generative Adversarial Network)은 비지도 학습에서도 컴퓨터 비젼의 문제들을 충분히 풀어나갈 수 있음을 보였고, 학습된 생성기를 활용하여 생성의 영역까지도 연구가 가능하게 하였다. GAN은 여러 가지 모델들과 결합하여 다양한 형태로 발전되고 있다. 기계학습에는 데이터 수집의 어려움이 있다. 너무 방대하면 노이즈를 제거를 통한 효과적인 데이터셋의 정제가 어렵고, 너무 작으면 작은 차이도 큰 노이즈가 되어 학습이 쉽지 않다. 본 논문에서는 GAN 모델에 영상 프레임 내의 얼굴 영역 추출을 위한 deep CNN 모델을 전처리 필터로 적용하여 두 사람의 제한된 수집데이터로 안정적으로 학습하여 다양한 표정의 합성 이미지를 만들어 낼 수 있는 방법을 제시하였다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권7호
• /
• pp.2547-2567
• /
• 2021

It has been widely acknowledged that occlusion impairments adversely distress many face recognition algorithms' performance. Therefore, it is crucial to solving the problem of face image occlusion in face recognition. To solve the image occlusion problem in face recognition, this paper aims to automatically de-occlude the human face majority or discriminative regions to improve face recognition performance. To achieve this, we decompose the generative process into two key stages and employ a separate generative adversarial network (GAN)-based network in both stages. The first stage generates an initial coarse face image without an occlusion mask. The second stage refines the result from the first stage by forcing it closer to real face images or ground truth. To increase the performance and minimize the artifacts in the generated result, a new refine loss (e.g., reconstruction loss, perceptual loss, and adversarial loss) is used to determine all differences between the generated de-occluded face image and ground truth. Furthermore, we build occluded face images and corresponding occlusion-free face images dataset. We trained our model on this new dataset and later tested it on real-world face images. The experiment results (qualitative and quantitative) and the comparative study confirm the robustness and effectiveness of the proposed work in removing challenging occlusion masks with various structures, sizes, shapes, types, and positions.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송∙미디어공학회 2018년도 하계학술대회
• /
• pp.347-348
• /
• 2018

본 논문에서는 야간 도로 영상을 보정하여 주간 영상으로 변환하는 알고리즘을 제안한다. 영상 변환 딥러닝 알고리즘인 Generative Adversarial Network(GAN)를 기반으로 주야간 도로 영상을 학습시켜 주야간 상호 변환이 가능한 시스템을 구현한다. 우선, 입력 영상에 대해 변환된 영상을 출력하는 generative network 를 정의한다. 또한, 변환된 영상을 다시 본래 영상으로 변환하는 inverse network 를 정의한다. Generative network 와 inverse network 를 모두 통과한 결과 영상과 본래 영상의 차 영상을 통해 손실 함수를 정의함으로써 파라미터를 목적에 맞게 학습시킬 수 있다. 또한, generative network 를 통과한 결과 영상과 목적하는 영상을 구분하는 discrimination network 를 정의하여 discrimination network 와 generative network 의 minimax two- player game 을 통해 변환된 영상이 실제 목적 영상과 유사하도록 유도한다. 제안하는 알고리즘을 적용하여 야간 도로 영상의 보정을 수행하면 주변 물체 인식이 어려운 야간 영상을 물체 인식이 용이한 주간 영상으로 변환 할 수 있다.

• KEPCO Journal on Electric Power and Energy
• /
• 제8권2호
• /
• pp.137-141
• /
• 2022

To inspect the power facility faults using artificial intelligence, it need that improve the accuracy of the diagnostic model are required. Data augmentation skill using generative adversarial network (GAN) is one of the best ways to improve deep learning performance. GAN model can create realistic-looking fake images using two competitive learning networks such as discriminator and generator. In this study, we intend to verify the effectiveness of virtual data generation technology by including the fake image of power facility generated through GAN in the deep learning training set. The GAN-based fake image was created for damage of LP insulator, and ResNet based normal and defect classification model was developed to verify the effect. Through this, we analyzed the model accuracy according to the ratio of normal and defective training data.