Mip-NeRF: A Multiscale Representation for Anti-Aliasing Neural Radiance Fields

被引:728
作者
Barron, Jonathan T. [1 ]
Mildenhall, Ben [1 ]
Tancik, Matthew [2 ]
Hedman, Peter [1 ]
Martin-Brualla, Ricardo [1 ]
Srinivasan, Pratul P. [1 ]
机构
[1] Google, Mountain View, CA 94043 USA
[2] Univ Calif Berkeley, Berkeley, CA 94720 USA
来源
2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021) | 2021年
关键词
D O I
10.1109/ICCV48922.2021.00580
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
The rendering procedure used by neural radiance fields (NeRF) samples a scene with a single ray per pixel and may therefore produce renderings that are excessively blurred or aliased when training or testing images observe scene content at different resolutions. The straightforward solution of supersampling by rendering with multiple rays per pixel is impractical for NeRF, because rendering each ray requires querying a multilayer perceptron hundreds of times. Our solution, which we call "mip-NeRF" (a la "mipmap"), extends NeRF to represent the scene at a continuously-valued scale. By efficiently rendering anti-aliased conical frustums instead of rays, mip-NeRF reduces objectionable aliasing artifacts and significantly improves NeRF's ability to represent fine details, while also being 7% faster than NeRF and half the size. Compared to NeRF, mip-NeRF reduces average error rates by 17% on the dataset presented with NeRF and by 60% on a challenging multiscale variant of that dataset that we present. Mip-NeRF is also able to match the accuracy of a brute-force supersampled NeRF on our multiscale dataset while being 22x faster.
引用
收藏
页码:5835 / 5844
页数:10
相关论文
共 53 条
[1]  
Amanatides John, 1984, SIGGRAPH
[2]  
[Anonymous], 2019, SIGGRAPH, DOI DOI 10.1145/3306346.3322980
[3]  
[Anonymous], 2019, CVPR, DOI DOI 10.1109/CVPR.2019.00254
[4]  
[Anonymous], 2007, NEURIPS
[5]  
[Anonymous], 2012, COMPUTER GRAPHICS FO
[6]   DERIVED CATEGORIES OF THE CAYLEY PLANE AND THE COADJOINT GRASSMANNIAN OF TYPE F [J].
Belmans, Pieter ;
Kuznetsov, Alexander ;
Smirnov, Maxim .
TRANSFORMATION GROUPS, 2023, 28 (01) :9-34
[7]  
Bi Sai, 2020, Neural reflectance fields for appearance acquisition
[8]  
Boss Mark, 2012, ARXIVCSCV201203918, V2020
[9]  
Bradbury J., 2018, JAX: composable transformations of Python+NumPy programs
[10]  
Brunet Dominique, 2011, IEEE TIP