http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
권주원(Juwon Kwon),윤길중(Gilly Yun),김승종(Seungjong Kim),정유채(Yuchae Jung),권순철(Soon Chul Kwon) 대한전자공학회 2020 대한전자공학회 학술대회 Vol.2020 No.8
In this paper, we described a technique for improving diagnostic performance using deep learning for Oligodendroglioma, one of brain tumors. Certain diseases, such as brain tumors, have a data imbalance problem because the number of patients in each hospital is not uniform. To solve this problem, we intended to introduce a data augmentation technique using GAN, a deep learning generation model. The experiment compared the training performance of the two groups with the dataset without fake data and the dataset with additional fake data. As a result, we showed a slight improvement in training performance and training stability.
SinGAN을 이용한 Paint to Image 스타일 변환 구현
박상언(Sangen Park),김준식(Junsik Kim),윤길중(Gilly Yun),권순철(Soon Chul Kwon) 대한전자공학회 2020 대한전자공학회 학술대회 Vol.2020 No.8
Recently, paint to image style transfer that converts a painting image into a photo-realistic image has been researched. This method requires hundreds of thousands to millions of datasets. The more datasets help you get the higher quality image, but it takes a lot of time to train. One way to solve this problem is using a single image as input. Approaches of previous single image GAN methods were limited to texture images and they were conditional. So, they still do not work well to learn the distribution of datasets made up of various classes. However, SinGAN uses a pyramid of fully convolutional GANs to learn the patch distribution at a different scale of the image, which can maintain the fine textures of the training image. Also, since it is unconditional, it can be applied not only to texture images, but also to general natural images. So even if you use a single image as input, you can get a photo-realistic image.