• International journal of advanced smart convergence
• /
• v.13 no.2
• /
• pp.195-204
• /
• 2024

In this paper, we investigate the comparative performance of two leading object detection architectures, YOLOv7 and Detection Transformer (DETR), across varying levels of data augmentation using CycleGAN. Our experiments focus on chest scan images within the context of biomedical informatics, specifically targeting the detection of abnormalities. The study reveals that YOLOv7 consistently outperforms DETR across all levels of augmented data, maintaining better performance even with 75% augmented data. Additionally, YOLOv7 demonstrates significantly faster convergence, requiring approximately 30 epochs compared to DETR's 300 epochs. These findings underscore the superiority of YOLOv7 for object detection tasks, especially in scenarios with limited data and when rapid convergence is essential. Our results provide valuable insights for researchers and practitioners in the field of computer vision, highlighting the effectiveness of YOLOv7 and the importance of data augmentation in improving model performance and efficiency.

• Information Systems Review
• /
• v.24 no.4
• /
• pp.97-118
• /
• 2022

As generative adversarial network (GAN) based oversampling techniques have achieved impressive results in class imbalance of unstructured dataset such as image, many studies have begun to apply it to solving the problem of imbalance in structured dataset. However, these studies have failed to reflect the characteristics of structured data due to changing the data structure into an unstructured data format. In order to overcome the limitation, this study adapted CycleGAN to reflect the characteristics of structured data, and proposed hybridization of synthetic minority oversampling technique (SMOTE) and the adapted CycleGAN. In particular, this study tried to overcome the limitations of existing studies by using a one-dimensional convolutional neural network unlike previous studies that used two-dimensional convolutional neural network. Oversampling based on the method proposed have been experimented using various datasets and compared the performance of the method with existing oversampling methods such as SMOTE and adaptive synthetic sampling (ADASYN). The results indicated the proposed hybrid oversampling method showed superior performance compared to the existing methods when data have more dimensions or higher degree of imbalance. This study implied that the classification performance of oversampling structured data can be improved using the proposed hybrid oversampling method that considers the characteristic of structured data.

• The Journal of Korea Robotics Society
• /
• v.13 no.1
• /
• pp.1-7
• /
• 2018

The image-to-image translation is one of the deep learning applications using image data. In this paper, we aim at improving the performance of object transfiguration which transforms a specific object in an image into another specific object. For object transfiguration, it is required to transform only the target object and maintain background images. In the existing results, however, it is observed that other parts in the image are also transformed. In this paper, we have focused on the structure of artificial neural networks that are frequently used in the existing methods and have improved the performance by adding constraints to the exiting structure. We also propose the advanced structure that combines the existing structures to maintain their advantages and complement their drawbacks. The effectiveness of the proposed methods are shown in experimental results.

• Journal of Korea Multimedia Society
• /
• v.22 no.1
• /
• pp.44-54
• /
• 2019

Recently, image-based object detection has made great progress with the introduction of Convolutional Neural Network (CNN). Many trials such as Region-based CNN, Fast R-CNN, and Faster R-CNN, have been proposed for achieving better performance in object detection. YOLO has showed the best performance under consideration of both accuracy and computational complexity. However, these data-driven detection methods including YOLO have the fundamental problem is that they can not guarantee the good performance without a large number of training database. In this paper, we propose a data sampling method using CycleGAN to solve this problem, which can convert styles while retaining the characteristics of a given input image. We will generate the insufficient data samples for training more robust object detection without efforts of collecting more database. We make extensive experimental results using the day-time and night-time road images and we validate the proposed method can improve the object detection accuracy of the night-time without training night-time object databases, because we converts the day-time training images into the synthesized night-time images and we train the detection model with the real day-time images and the synthesized night-time images.

• The Journal of Bigdata
• /
• v.7 no.1
• /
• pp.173-185
• /
• 2022

Nowadays, Korean webtoons are leading the global digital comic market. Webtoons are being serviced in various languages around the world, and dramas or movies produced with Webtoons' IP (Intellectual Property Rights) have become a big hit, and more and more webtoons are being visualized. However, with the success of these webtoons, the working environment of webtoon creators is emerging as an important issue. According to the 2021 Cartoon User Survey, webtoon creators spend 10.5 hours a day on creative activities on average. Creators have to draw large amount of pictures every week, and competition among webtoons is getting fiercer, and the amount of paintings that creators have to draw per episode is increasing. Therefore, this study proposes to generate webtoon background images using deep learning algorithms and use them for webtoon production. The main character in webtoon is an area that needs much of the originality of the creator, but the background picture is relatively repetitive and does not require originality, so it can be useful for webtoon production if it can create a background picture similar to the creator's drawing style. Background generation uses CycleGAN, which shows good performance in image-to-image translation, and CartoonGAN, which is specialized in the Cartoon style image generation. This deep learning-based image generation is expected to shorten the working hours of creators in an excessive work environment and contribute to the convergence of webtoons and technologies.

• Journal of Ocean Engineering and Technology
• /
• v.34 no.6
• /
• pp.442-453
• /
• 2020

Underwater color images suffer from low visibility and color cast effects caused by light attenuation by water and floating particles. This study applied single image enhancement techniques to enhance the quality of underwater images and compared their performance with real underwater images taken in Korean waters. Dark channel prior (DCP), gradient transform, image fusion, and generative adversarial networks (GAN), such as cycleGAN and underwater GAN (UGAN), were considered for single image enhancement. Their performance was evaluated in terms of underwater image quality measure, underwater color image quality evaluation, gray-world assumption, and blur metric. The DCP saturated the underwater images to a specific greenish or bluish color tone and reduced the brightness of the background signal. The gradient transform method with two transmission maps were sensitive to the light source and highlighted the region exposed to light. Although image fusion enabled reasonable color correction, the object details were lost due to the last fusion step. CycleGAN corrected overall color tone relatively well but generated artifacts in the background. UGAN showed good visual quality and obtained the highest scores against all figures of merit (FOMs) by compensating for the colors and visibility compared to the other single enhancement methods.

• Journal of the Korean Society of Radiology
• /
• v.18 no.1
• /
• pp.37-44
• /
• 2024

Even though MR can reveal excellent soft-tissue contrast and functional information, CT is also required for electron density information for accurate dose calculation in Radiotherapy. For the fusion of MRI and CT images in RT treatment planning workflow, patients are normally scanned on both MRI and CT imaging modalities. Recently deep-learning-based generations of CT images from MR images became possible owing to machine learning technology. This eliminated CT scanning work. This study implemented a CycleGan deep-learning-based CT image generation from MR images. Three CT generators whose learning is based on T1- , T2- , or T1-&T2-weighted MR images were created, respectively. We found that the T1-weighted MR image-based generator can generate better than other CT generators when T1-weighted MR images are input. In contrast, a T2-weighted MR image-based generator can generate better than other CT generators do when T2-weighted MR images are input. The results say that the CT generator from MR images is just outside the practical clinics and the specific weight MR image-based machine-learning generator can generate better CT images than other sequence MR image-based generators do.

• Geophysics and Geophysical Exploration
• /
• v.26 no.2
• /
• pp.37-51
• /
• 2023

The ground roll is the most common coherent noise in land seismic data and has an amplitude much larger than the reflection event we usually want to obtain. Therefore, ground roll suppression is a crucial step in seismic data processing. Several techniques, such as f-k filtering and curvelet transform, have been developed to suppress the ground roll. However, the existing methods still require improvements in suppression performance and efficiency. Various studies on the suppression of ground roll in seismic data have recently been conducted using deep learning methods developed for image processing. In this paper, we introduce three models (DnCNN (De-noiseCNN), pix2pix, and CycleGAN), based on convolutional neural network (CNN) or conditional generative adversarial network (cGAN), for ground roll suppression and explain them in detail through numerical examples. Common shot gathers from the same field were divided into training and test datasets to compare the algorithms. We trained the models using the training data and evaluated their performances using the test data. When training these models with field data, ground roll removed data are required; therefore, the ground roll is suppressed by f-k filtering and used as the ground-truth data. To evaluate the performance of the deep learning models and compare the training results, we utilized quantitative indicators such as the correlation coefficient and structural similarity index measure (SSIM) based on the similarity to the ground-truth data. The DnCNN model exhibited the best performance, and we confirmed that other models could also be applied to suppress the ground roll.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.07a
• /
• pp.179-181
• /
• 2023

본 논문에서는 GAN 기반의 영상 생성 방법론을 이용해 delayed PET 영상을 생성하는 연구를 수행하였다. PET은 양전자를 방출하는 방사성 동위원소를 표지한 방사성의약품의 체내 분포를 시각화함으로서 암 세포 진단에 이용되는 의료영상 기법이다. 하지만 PET의 스캔 과정에서 방사성의약품이 체내에 분포하는 데에 걸리는 시간이 오래 걸린다는 문제점이 존재한다. 따라서 본 연구에서는 방사성의약품이 충분히 분포되지 않은 상태에서 얻은 PET 영상을 통해 목표로 하는 충분히 시간이 지난 후에 얻은 PET 영상을 생성하는 모델을 GAN (generative adversarial network)에 기반한 image-to-image translation(I2I)를 통해 수행했다. 특히, 생성 전후의 영상 간의 영상 쌍을 고려한 paired I2I인 Pix2pix와 이를 고려하지 않은 unpaired I2I인 CycleGAN 두 가지의 방법론을 비교하였다. 연구 결과, Pix2pix에 기반해 생성한 delayed PET 영상이 CycleGAN을 통해 생성한 영상에 비해 영상 품질이 좋음을 확인했으며, 또한 실제 획득한 ground-truth delayed PET 영상과의 유사도 또한 더 높음을 확인할 수 있었다. 결과적으로, 딥러닝에 기반해 early PET을 통해 delayed PET을 생성할 수 있었으며, paired I2I를 적용할 경우 보다 높은 성능을 기대할 수 있었다. 이를 통해 PET 영상 획득 과정에서 방사성의약품의 체내 분포에 소요되는 시간을 딥러닝 모델을 통해 줄여 PET 이미징 과정의 시간적 비용을 절감하는 데에 크게 기여할 수 있을 것으로 기대된다.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.44 no.3
• /
• pp.397-405
• /
• 2024

In order to improve the visibility of nighttime thermal infrared images, a simulation method with daytime color images was proposed. As a simulation method consisting of two steps, the daytime thermal infrared image was simulated by learning the unpaired nighttime thermal infrared image and daytime thermal infrared image, then the result was translated into a daytime color image. A temperature change regression equation was constructed and applied to reflect the systematic characteristics of temperature changes in daytime and nighttime images, and day and night simulation and colorization were trained and modeled by CycleGAN. For the experimental area, 100 images were captured and used for training. As a result, the simulation showed an average SSIM of 0.2449 and a PSNR of 51.2254. It was confirmed that the method could simulate complex and detailed features such as vegetation.