• Smart Media Journal
• /
• v.10 no.1
• /
• pp.79-87
• /
• 2021

In recent years research on automatic font generation with machine learning mainly focus on using transformation-based methods, in comparison, generative model-based methods of font generation have received less attention. Transformation-based methods learn a mapping of the transformations from an existing input to a target. This makes them ambiguous because in some cases a single input reference may correspond to multiple possible outputs. In this work, we focus on font generation using the generative model-based methods which learn the buildup of the characters from noise-to-image. We propose a novel way to train a conditional generative deep neural model so that we can achieve font style control on the generated font images. Our research demonstrates how to generate new font images conditioned on both character class labels and character style labels when using the generative model-based methods. We achieve this by introducing a modified generator network which is given inputs noise, character class, and style, which help us to calculate losses separately for the character class labels and character style labels. We show that adding the character style vector on top of the character class vector separately gives the model rich information about the font and enables us to explicitly specify not only the character class but also the character style that we want the model to generate.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.11
• /
• pp.5427-5445
• /
• 2019

While deep neural networks have achieved remarkable performance in representation learning, a huge amount of labeled training data are usually required by supervised deep models such as convolutional neural networks. In this paper, we propose a new representation learning method, namely generative adversarial networks (GAN) based bagging deep convolutional autoencoders (GAN-BDCAE), which can map data to diverse hierarchical representations in an unsupervised fashion. To boost the size of training data, to train deep model and to aggregate diverse learning machines are the three principal avenues towards increasing the capabilities of representation learning of neural networks. We focus on combining those three techniques. To this aim, we adopt GAN for realistic unlabeled sample generation and bagging deep convolutional autoencoders (BDCAE) for robust feature learning. The proposed method improves the discriminative ability of learned feature embedding for solving subsequent pattern recognition problems. We evaluate our approach on three standard benchmarks and demonstrate the superiority of the proposed method compared to traditional unsupervised learning methods.

• Smart Media Journal
• /
• v.8 no.4
• /
• pp.33-37
• /
• 2019

Because the ECG signals are time-series data acquired as time elapses, it is important to obtain comparative data the same in size as the enrolled data every time. This paper suggests a network model of GAN (Generative Adversarial Networks) based on an auxiliary classifier to generate synthetic ECG signals which may address the different data size issues. The Cosine similarity and Cross-correlation are used to examine the similarity of synthetic ECG signals. The analysis shows that the Average Cosine similarity was 0.991 and the Average Euclidean distance similarity based on cross-correlation was 0.25: such results indicate that data size difference issue can be resolved while the generated synthetic ECG signals, similar to real ECG signals, can create synthetic data even when the registered data are not the same as the comparative data in size.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.1
• /
• pp.9-18
• /
• 2021

Although Generative Adversarial Networks (GANs) have gained great popularity in computer vision and related fields, generating audio signals independently has yet to be presented. Unlike images, an audio signal is a sampled signal consisting of discrete samples, so it is not easy to learn the signals using CNN architectures, which is widely used in image generation tasks. In order to overcome this difficulty, GAN researchers proposed a strategy of applying time-frequency representations of audio to existing image-generating GANs. Following this strategy, we propose an improved method for increasing the fidelity of synthesized audio signals generated by using GANs. Our method is demonstrated on a public speech dataset, and evaluated by Fréchet Inception Distance (FID). When employing our method, the FID showed 10.504, but 11.973 as for the existing state of the art method (lower FID indicates better fidelity).

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.11a
• /
• pp.565-567
• /
• 2022

질병을 앓고 있는 환자는 상태에 따라 병실, 주거지, 요양원 등 조호환경 내 생활 시 의료 인력의 지속적인 추적 및 관찰을 통해 신체에 이상이 생긴 경우 이를 감지하고, 신속하게 조치할 수 있도록 해야 한다. 의료 인력이 직접 환자를 확인하는 방법은 의료 인력의 반복적인 노동이 요구되며 실시간으로 환자를 확인해야 한다는 특성상 의료 인력이 상주해야 하기에 이는 곧, 의료 인력의 부족과 낭비로 이어진다. 해당 문제 해결을 위해 의료 인력을 대신하여 조호환경 내 환자의 상태를 실시간으로 모니터링할 수 있는 딥러닝 모델들이 연구되고 있다. 딥러닝 모델은 데이터의 수가 많을수록 강인한 모델을 설계할 수 있으며, 데이터셋의 배경, 객체의 특징 분포 등 다양한 조건에 영향을 받기 때문에 학습에 필요한 도메인을 가지는 많은 양의 전처리된 데이터를 수집해야 한다. 따라서, 조호환경 내 환자에 대한 데이터셋이 필요하지만, 공개된 데이터셋의 경우 양이 매우 적으며 이를 반전, 회전기법 등을이용할 경우 데이터의 수를 늘릴 수 있지만, 같은 분포의 특징을 가지는 데이터가 생성되기에 데이터 증강 기법을 단순하게 적용하면 딥러닝 모델의 과적합을 야기한다. 또한, 조호환경 내 이미지 데이터셋은 얼굴 노출과 같은 개인정보가 포함 될 수 있으며 이를 보호하기 위해 정보들을 비식별화 해야 한다는 문제점이 있다. 따라서 본 논문에서는 조호환경에서 수집된 데이터 증강을 위한 Style-Generative Adversarial Networks 기법을 적용하여 조호환경 데이터셋 수집에 효과적인 증강 기법을 제안한다.

• Journal of Platform Technology
• /
• v.11 no.6
• /
• pp.34-46
• /
• 2023

The design of non-pneumatic tires, which are created by filling the space between the wheel and the tread with elastomeric compounds or polygonal spokes, has become an important research topic in the automotive and aerospace industries. In this study, a system was designed for the design of non-pneumatic tires through the implementation of a generative adversarial network. We specifically examined factors that could impact the design, including the type of non-pneumatic tire, its intended usage environment, manufacturing techniques, distinctions from pneumatic tires, and how spoke design affects load distribution. Using OpenCV, various shapes and spoke configurations were generated as images, and a GAN model was trained on the projected GANs to generate shapes and spokes for non-pneumatic tire designs. The designed non-pneumatic tires were labeled as available or not, and a Vision Transformer image classification AI model was trained on these labels for classification purposes. Evaluation of the classification model show convergence to a near-zero loss and a 99% accuracy rate confirming the generation of non-pneumatic tire designs.

• Nuclear Engineering and Technology
• /
• v.56 no.8
• /
• pp.2958-2973
• /
• 2024

Sliding bearings are crucial rotating mechanical components in nuclear power plants, and their failures can result in severe economic losses and human casualties. Deep learning provides a new approach to bearing fault diagnosis, but there is currently a lack of a universal fault diagnosis model for studying bearing-rotor systems under various operating conditions, speeds and faults. Research on bearing-rotor systems supported by sliding bearings is limited, leading to insufficient fault data. To address these issues, this paper proposes a fault diagnosis model framework for bearing-rotor systems based on a deep convolutional generative adversarial network (TF-DLGAN). This model not only exhibits outstanding fault diagnosis performance but also addresses the issue of insufficient fault data. An experimental platform is constructed to conduct fault experiments under various operating conditions, speeds and faults, establishing a dataset for sliding bearing-rotor system faults. Finally, the model's effectiveness is validated using this dataset.

• Journal of the Korea Computer Graphics Society
• /
• v.24 no.2
• /
• pp.1-9
• /
• 2018

In this paper, we propose a method to restore corrupted black and white facial images to color. Previous studies have shown that when coloring damaged black and white photographs, such as old ID photographs, the area around the damaged area is often incorrectly colored. To solve this problem, this paper proposes a method of restoring the damaged area of input photo first and then performing colorization based on the result. The proposed method consists of two steps: BEGAN (Boundary Equivalent Generative Adversarial Networks) model based restoration and CNN (Convolutional Neural Network) based coloring. Our method uses the BEGAN model, which enables a clearer and higher resolution image restoration than the existing methods using the DCGAN (Deep Convolutional Generative Adversarial Networks) model for image restoration, and performs colorization based on the restored black and white image. Finally, we confirmed that the experimental results of various types of facial images and masks can show realistic color restoration results in many cases compared with the previous studies.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.8
• /
• pp.3942-3961
• /
• 2019

Recently, the models of deep super-resolution networks can successfully learn the non-linear mapping from the low-resolution inputs to high-resolution outputs. However, for large scaling factors, this approach has difficulties in learning the relation of low-resolution to high-resolution images, which lead to the poor restoration. In this paper, we propose Stage Generative Adversarial Networks (Stage-GAN) with semantic maps for image super-resolution (SR) in large scaling factors. We decompose the task of image super-resolution into a novel semantic map based reconstruction and refinement process. In the initial stage, the semantic maps based on the given low-resolution images can be generated by Stage-0 GAN. In the next stage, the generated semantic maps from Stage-0 and corresponding low-resolution images can be used to yield high-resolution images by Stage-1 GAN. In order to remove the reconstruction artifacts and blurs for high-resolution images, Stage-2 GAN based post-processing module is proposed in the last stage, which can reconstruct high-resolution images with photo-realistic details. Extensive experiments and comparisons with other SR methods demonstrate that our proposed method can restore photo-realistic images with visual improvements. For scale factor ${\times}8$, our method performs favorably against other methods in terms of gradients similarity.

• Korean Journal of Remote Sensing
• /
• v.38 no.6_1
• /
• pp.1357-1369
• /
• 2022

Cloud removal is an essential image processing step for any task requiring time-series optical images, such as vegetation monitoring and change detection. This paper presents a two-stage cloud removal method that combines conditional generative adversarial networks (cGANs) with regression-based calibration to construct a cloud-free time-series optical image set. In the first stage, the cGANs generate initial prediction results using quantitative relationships between optical and synthetic aperture radar images. In the second stage, the relationships between the predicted results and the actual values in non-cloud areas are first quantified via random forest-based regression modeling and then used to calibrate the cGAN-based prediction results. The potential of the proposed method was evaluated from a cloud removal experiment using Sentinel-2 and COSMO-SkyMed images in the rice field cultivation area of Gimje. The cGAN model could effectively predict the reflectance values in the cloud-contaminated rice fields where severe changes in physical surface conditions happened. Moreover, the regression-based calibration in the second stage could improve the prediction accuracy, compared with a regression-based cloud removal method using a supplementary image that is temporally distant from the target image. These experimental results indicate that the proposed method can be effectively applied to restore cloud-contaminated areas when cloud-free optical images are unavailable for environmental monitoring.