지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
논문 기본 정보
- 자료유형
- 학위논문
- 저자정보
- 지도교수
- 조근식
- 발행연도
- 2020
- 저작권
- 인하대학교 논문은 저작권에 의해 보호받습니다.
이용수1
이 논문의 연구 히스토리 (2)
- 이 논문의 후속연구가 궁금하신가요?
- 연관 학술논문 또는 학술발표를 통해 보다 발전된 연구결과를 확인하실 수 있습니다.
- 이 논문의 연구 히스토리 확인하기
초록· 키워드
오류제보하기
Generative Adversarial Networks (GANs) made a huge contribution to the development of content creation technologies. An important place in this advancement takes video generation due to the need for human animation applications, automatic trailer or movie generation. Therefore, taking advantage of various GANs, we proposed own method for Human Action Video Generation called GOHAG: GANs Orchestration for Human Actions Generation. GOHAG is a combination of three GANs, where Poses generation GAN (PGAN) creates a sequence of poses, Poses Optimization GAN (POGAN) optimizes them, and Frames generation GAN (FGAN) attaches texture for the sequence, creating a video.
Extensive experiments were conducted on a public dataset and compared with existing work in the field of video generation. The evaluation showed that GOHAG is outperforming the recent technique in various metrics. In addition, a private dataset of various human actions was collected to widen the experiments, where GOHAG performed better to compare with the public dataset. The proposed technique proved to have the ability to produce a smooth and plausible video of high-quality, where the human poses are created and being used to control the texturing phase.
Extensive experiments were conducted on a public dataset and compared with existing work in the field of video generation. The evaluation showed that GOHAG is outperforming the recent technique in various metrics. In addition, a private dataset of various human actions was collected to widen the experiments, where GOHAG performed better to compare with the public dataset. The proposed technique proved to have the ability to produce a smooth and plausible video of high-quality, where the human poses are created and being used to control the texturing phase.
목차
Chapter 1 Introduction 6Chapter 2 Related Works 9Chapter 3 Human Action Video Generation using GANs 113.1 Action Representation 123.2 Pose Sequence Generation 13Poses Generator GAN 13Poses Optimization GAN 163.3 Frames Sequence Generation. 18Frames Generator GAN 183.4 Implementation Details 20Chapter 4 Experimental Results 214.1 Datasets 224.2 Quantitative Evaluation 23L2 Distance 24Video Quality Evaluation 26Inception Score and Frechet Inception Distance 284.3 Qualitative Evaluation 29User Evaluation. 29Chapter 5 Conclusions and Future Work 32References 33Acknowledgement 35
최근 본 자료
댓글(0)
0