• Smart Media Journal
• /
• v.8 no.2
• /
• pp.46-57
• /
• 2019

In recent research, deep learning-based methods have achieved state-of-the-art performance in various computer vision tasks. However, these methods are commonly supervised, and require huge amounts of annotated data to train. Acquisition of data demands an additional costly effort, particularly for the tasks where it becomes challenging to obtain large amounts of data considering the time constraints and the requirement of professional human diligence. In this paper, we present a data level synthetic sampling solution to learn from small and imbalanced data sets using Generative Adversarial Networks (GANs). The reason for using GANs are the challenges posed in various fields to manage with the small datasets and fluctuating amounts of samples per class. As a result, we present an approach that can improve learning with respect to data distributions, reducing the partiality introduced by class imbalance and hence shifting the classification decision boundary towards more accurate results. Our novel method is demonstrated on a small dataset of 2789 tomato plant disease images, highly corrupted with class imbalance in 9 disease categories. Moreover, we evaluate our results in terms of different metrics and compare the quality of these results for distinct classes.

• 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.

• KIPS Transactions on Software and Data Engineering
• /
• v.8 no.6
• /
• pp.257-264
• /
• 2019

Recently, deep neural network based approaches have shown a good performance for various fields of natural language processing. A huge amount of training data is essential for building a deep neural network model. However, collecting a large size of training data is a costly and time-consuming job. A data augmentation is one of the solutions to this problem. The data augmentation of text data is more difficult than that of image data because texts consist of tokens with discrete values. Generative adversarial networks (GANs) are widely used for image generation. In this work, we generate sentimental texts by using one of the GANs, CS-GAN model that has a discriminator as well as a classifier. We evaluate the usefulness of generated sentimental texts according to various measurements. CS-GAN model not only can generate texts with more diversity but also can improve the performance of its classifier.

• Journal of Korean Neurosurgical Society
• /
• v.63 no.3
• /
• pp.386-396
• /
• 2020

Objective : To generate synthetic spine magnetic resonance (MR) images from spine computed tomography (CT) using generative adversarial networks (GANs), as well as to determine the similarities between synthesized and real MR images. Methods : GANs were trained to transform spine CT image slices into spine magnetic resonance T2 weighted (MRT2) axial image slices by combining adversarial loss and voxel-wise loss. Experiments were performed using 280 pairs of lumbar spine CT scans and MRT2 images. The MRT2 images were then synthesized from 15 other spine CT scans. To evaluate whether the synthetic MR images were realistic, two radiologists, two spine surgeons, and two residents blindly classified the real and synthetic MRT2 images. Two experienced radiologists then evaluated the similarities between subdivisions of the real and synthetic MRT2 images. Quantitative analysis of the synthetic MRT2 images was performed using the mean absolute error (MAE) and peak signal-to-noise ratio (PSNR). Results : The mean overall similarity of the synthetic MRT2 images evaluated by radiologists was 80.2%. In the blind classification of the real MRT2 images, the failure rate ranged from 0% to 40%. The MAE value of each image ranged from 13.75 to 34.24 pixels (mean, 21.19 pixels), and the PSNR of each image ranged from 61.96 to 68.16 dB (mean, 64.92 dB). Conclusion : This was the first study to apply GANs to synthesize spine MR images from CT images. Despite the small dataset of 280 pairs, the synthetic MR images were relatively well implemented. Synthesis of medical images using GANs is a new paradigm of artificial intelligence application in medical imaging. We expect that synthesis of MR images from spine CT images using GANs will improve the diagnostic usefulness of CT. To better inform the clinical applications of this technique, further studies are needed involving a large dataset, a variety of pathologies, and other MR sequence of the lumbar spine.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.4
• /
• pp.1166-1186
• /
• 2022

Artistic font design has become an integral part of visual media. However, without prior knowledge of the font domain, it is difficult to create distinct font styles. When the number of characters is limited, this task becomes easier (e.g., only Latin characters). However, designing CJK (Chinese, Japanese, and Korean) characters presents a challenge due to the large number of character sets and complexity of the glyph components in these languages. Numerous studies have been conducted on automating the font design process using generative adversarial networks (GANs). Existing methods rely heavily on reference fonts and perform font style conversions between different fonts. Additionally, rather than capturing style information for a target font via multiple style images, most methods do so via a single font image. In this paper, we propose a network architecture for generating multilingual font sets that makes use of geometric structures as content. Additionally, to acquire sufficient style information, we employ multiple style images belonging to a single font style simultaneously to extract global font style-specific information. By utilizing the geometric structural information of content and a few stylized images, our model can generate an entire font set while maintaining the style. Extensive experiments were conducted to demonstrate the proposed model's superiority over several baseline methods. Additionally, we conducted ablation studies to validate our proposed network architecture.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2021.06a
• /
• pp.105-108
• /
• 2021

Generative adversarial networks (GANs) have reached a great result at creating the synthesis image, especially in the face generation task. Unlike other deep learning tasks, the input of GANs is usually the random vector sampled by a probability distribution, which leads to unstable training and unpredictable output. One way to solve those problems is to employ the label condition in both the generator and discriminator. CelebA and FFHQ are the two most famous datasets for face image generation. While CelebA contains attribute annotations for more than 200,000 images, FFHQ does not have attribute annotations. Thus, in this work, we introduce a method to learn the attributes from CelebA then predict both soft and hard labels for FFHQ. The evaluated result from our model achieves 0.7611 points of the metric is the area under the receiver operating characteristic curve.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.7
• /
• pp.1794-1806
• /
• 2023

This study presents a method for capturing photographs of users as input and converting them into 2D character animation sprites using a generative adversarial network-based artificial intelligence network. Traditionally, 2D character animations have been created by manually creating an entire sequence of sprite images, which incurs high development costs. To address this issue, this study proposes a technique that combines motion videos and sample 2D images. In the 2D sprite generation process that uses the proposed technique, a sequence of images is extracted from real-life images captured by the user, and these are combined with character images from within the game. Our research aims to leverage cutting-edge deep learning-based image manipulation techniques, such as the GAN-based motion transfer network (impersonator) and background noise removal (U² -Net), to generate a sequence of animation sprites from a single image. The proposed technique enables the creation of diverse animations and motions just one image. By utilizing these advancements, we focus on enhancing productivity in the game and animation industry through improved efficiency and streamlined production processes. By employing state-of-the-art techniques, our research enables the generation of 2D sprite images with various motions, offering significant potential for boosting productivity and creativity in the industry.

• Proceedings of the Korea Contents Association Conference
• /
• 2018.05a
• /
• pp.459-460
• /
• 2018

In this paper, we present a data augmentation method that generates synthetic plant disease images using Generative Adversarial Networks (GANs). We propose a training scheme that first uses classical data augmentation techniques to enlarge the training set and then further enlarges the data size and its diversity by applying GAN techniques for synthetic data augmentation. Our method is demonstrated on a limited dataset of 2789 images of tomato plant diseases (Gray mold, Canker, Leaf mold, Plague, Leaf miner, Whitefly etc.).

• Electronics and Telecommunications Trends
• /
• v.34 no.3
• /
• pp.34-42
• /
• 2019

This study introduces artificial intelligence (AI) techniques for video generation. For an effective illustration, techniques for video generation are classified as either semi-automatic or automatic. First, we discuss some recent achievements in semi-automatic video generation, and explain which types of AI techniques can be applied to produce films and improve film quality. Additionally, we provide an example of video content that has been generated by using AI techniques. Then, two automatic video-generation techniques are introduced with technical details. As there is currently no feasible automatic video-generation technique that can generate commercial videos, in this study, we explain their technical details, and suggest the future direction for researchers. Finally, we discuss several considerations for more practical automatic video-generation techniques.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.07a
• /
• pp.145-147
• /
• 2022

본 논문에서는 훼손되어 식별할 수 없는 탑 이미지를 비롯해 낮은 해상도의 탑 이미지를 개선하기 위해 우리는 탑 이미지의 화질 개선을 인공지능을 이용하여 빠르게 개선을 해 보고자 한다. 최근에 Generative Adversarial Networks(GANS) 알고리즘에서 SrGAN 알고리즘이 나오면서 이미지 생성, 이미지 복원, 해상도 변화 분야가 지속해서 발전하고 있다. 이에 본 연구에서는 다양한 GAN 알고리즘을 화질 개선에 적용해 보았다. 탑 이미지에 GAN 알고리즘 중 SrGan을 적용하였으며 실험한 결과 Srgan 알고리즘은 학습이 진행되었으며, 낮은 해상도의 탑 이미지가 높은 해상도, 초고해상도 이미지가 생성되는 것을 확인했다.