사람의 실루엣 이미지 기반 3D 메시 데이터 생성 기술 개발

Abstract

3D human modeling has become a vital research topic for providing user-friendly interfaces in various application areas. In this paper, we propose a new network for succinctly generating a 3D model of the human body based on orthogonal silhouettes. Utilizing the CAESAR dataset, we sample the contour of the silhouette and train the mapping between the silhouette and the 3D model through a neural network. We have expanded the number of training data through data augmentation, resulting in a 3.41% reduction in the average distance error between vertices. This approach paves the way for potential applications in fields such as virtual fitting, custom design, and body health monitoring through 3D human modeling.