간략화 메쉬의 품질 향상을 위한 역전파 기반 최적화 방법 (Backpropagation-Based Optimization Method for Quality Improvement of Simplified Meshes)

서지정보

· 발행기관 : 한국CDE학회
· 수록지 정보 : 한국CDE학회 논문집 / 28권 / 4호 / 398 ~ 407페이지
· 저자명 : 정동원, 강봉구, 권순조, 김형기

초록

Triangle mesh is a common method in 3D computer graphics for representing an object's 3D shape using vertex and triangle connectivity information on its surface. However, representing the shape in high resolution often requires many triangles, resulting in increased computational complexity and storage requirements. Mesh simplification aims to reduce the number of vertices and triangles while minimizing shape information loss between the original mesh and the simplified mesh. Existing methods typically rely on algorithms that iteratively remove vertices or edges, potentially damaging shape information in specific areas, which varies depending on the simplification criteria. In this paper, we introduce a backpropagation-based optimization method to restore damaged shape information during mesh simplification. By setting the simplified mesh as the initial mesh and optimizing the offset vector for mesh transformation, we minimize the distance between the sampled point cloud from the deformed mesh and the original mesh. We conducted experiments on the TOSCA dataset and compared the shape similarity between the results from QEM and ours. Results showed up to 6.36%, 7.7%, and 54.56% improvements in Chamfer distance, forward and backward point cloud-mesh distances, respectively. Our experiments demonstrate that our approach effectively restores shape information without altering the number of triangles.

영어초록

Triangle mesh is a common method in 3D computer graphics for representing an object's 3D shape using vertex and triangle connectivity information on its surface. However, representing the shape in high resolution often requires many triangles, resulting in increased computational complexity and storage requirements. Mesh simplification aims to reduce the number of vertices and triangles while minimizing shape information loss between the original mesh and the simplified mesh. Existing methods typically rely on algorithms that iteratively remove vertices or edges, potentially damaging shape information in specific areas, which varies depending on the simplification criteria. In this paper, we introduce a backpropagation-based optimization method to restore damaged shape information during mesh simplification. By setting the simplified mesh as the initial mesh and optimizing the offset vector for mesh transformation, we minimize the distance between the sampled point cloud from the deformed mesh and the original mesh. We conducted experiments on the TOSCA dataset and compared the shape similarity between the results from QEM and ours. Results showed up to 6.36%, 7.7%, and 54.56% improvements in Chamfer distance, forward and backward point cloud-mesh distances, respectively. Our experiments demonstrate that our approach effectively restores shape information without altering the number of triangles.

참고자료

· 없음