Generative adversarial nets를 이용한 빈센트 반 고흐 이미지 생성 시스템

• 영문명: Vincent Van Gogh image generation system using generative adversarial nets Focusing on the Gan Method
• 발행기관: 한국디자인리서치학회
• 저자명: 조수영(Cho, Sooyoung) 손채봉(Sohn, Chae-Bong) 김문석(Kim, Moon Seok)
• 간행물 정보: 『한국디자인리서치』Vol2, No. 3(통권 6권), 85~92쪽, 전체 8쪽
• 주제분류: 예술체육 > 예술일반
• 파일형태: PDF
• 발행일자: 2017.12.31

영문 초록

CNN and RNN are classifiers for image and speech recognition, and are used in many computer vision. However, this model alone does not produce images or voices. So we used GAN (Generative Adversarial Nets), which is a non-biped learning, to create and restore objects. To determine the distribution of the assumed model data, we use the generative model G and the Discriminator model D to judge the probability of each case in order to distinguish whether the sample came from the training data from the actual or model G Respectively. This is called minimax two -player game. The model G makes a mistake maximize in order to make a mistake, and the model D makes a mistake minimize in order not to make a mistake. Model D uses technology such as recognition technology, which is a conventional prayer learning method. Through this technique, the damaged or modified objects are compared with the training sample to classify and confirm what kind of object is the first type. There is no need for another network to generate the sample through training and it will be evaluated and confirmed through experiments.

목차

1. 서론

2. 이론적 배경

3. GAN의 활용 사례

4. 실험

5. 결론

키워드

Generative Adversarial Nets CNN DCGANs