[3DGS] Mip-Splatting: Alias-free 3D Gaussian Splatting / Ver.Kor
2025. 3. 31. 10:21
Paper Information
- Title : Mip-Splatting: Alias-free 3D Gaussian Splatting
- Journal : CVPR 2024
- Author : Yu, Zehao, et al.
https://niujinshuchong.github.io/mip-splatting/
Mip-Splatting
Recently, 3D Gaussian Splatting (3DGS) has demonstrated impressive novel view synthesis results, reaching high fidelity and efficiency. However, strong artifacts can be observed when changing the sampling rate, e.g., by changing focal length or camera dist
niujinshuchong.github.io
Abstract
Problem :
strong artifacts observed when changing the sampling rate
=> cause : lack of 3D frequency constrainnts and the usage of a 2D dilation filter
Method : Mip Splatting
Details :
- 3D smoothing filter
- 2D Mip Filter (replace 2D dilation with 2D Mip filter)
Results :
- competitive with SOTA methods
- outperform SOTA methods in out-of-distributionn scenarios
'Computer Science > AI' 카테고리의 다른 글
