一种联合采样的神经网络光场

doi:10.13190/j.jbupt.2021-095

北京邮电大学学报 ›› 2021, Vol. 44 ›› Issue (6): 109-115.doi: 10.13190/j.jbupt.2021-095

一种联合采样的神经网络光场

刘绍华¹, 李明豪^1,2, 李兆歆², 毛天露², 刘京³

1. 北京邮电大学信息物理融合系统研究实验室, 北京 100876;
2. 中国科学院计算技术研究所, 北京 100190;
3. 河北师范大学软件学院, 石家庄 050024

收稿日期:2021-05-23 出版日期:2021-12-28 发布日期:2021-12-28
通讯作者: 李兆歆(1983—),男,助理研究员,E-mail:cszli@hotmail.com. E-mail:cszli@hotmail.com
作者简介:刘绍华(1976—),男,副教授,博士生导师.
基金资助:
国家自然科学基金项目(91938301,62172392)

Neural Radiance Field by Joint Sampling

LIU Shao-hua¹, LI Ming-hao^1,2, LI Zhao-xin², MAO Tian-lu², LIU Jing³

1. Cyber-Physical Systems Laboratory, Beijing University of Posts and Telecommunications, Beijing 100876, China;
2. Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China;
3. School of Software, Hebei Normal University, Shijiazhuang 050024, China

Received:2021-05-23 Online:2021-12-28 Published:2021-12-28

摘要/Abstract

摘要： 相比传统的光场绘制技术,神经网络光场(NeRF)方法可使用神经网络拟合场景的光线采样,将隐式编码输入图片的光场,合成新视图. 针对NeRF方法训练时间长,绘制视图慢的问题,提出了一种基于联合采样的NeRF方法,通过使粗糙网络和细腻网络共享均匀采样结果的方法,减少了不必要的光线采样,从而加快了网络训练和视图合成的速度. 实验结果表明,在取得近似视图合成质量的情况下,与NeRF方法相比,所提方法的训练时间减少了20%,视图合成的效率提高了25%.

关键词: 神经网络光场, 光线采样, 联合采样, 视图合成

Abstract: Compared with the traditional light field, the neural reflectance field (NeRF) method uses the neural network to fit the light sampling of scenes, which implicitly encodes the light field from input images to render novel view. However, NeRF method requires long training time and has slow rendering speed. To solve this problem, a joint sampling-based NeRF is proposed, which can make the coarse network and fine network share uniform sampling results, thereby accelerating the network training and view synthesis by reducing unnecessary light sampling. The experiments demonstrate that, in the case of the similar view synthesis quality, compared with the baseline method, the proposed method can reduce the training time by 20% and improve the view synthesis efficiency by 25%.

Key words: neural reflectance field, light sampling, joint sampling, view synthesis

中图分类号:

TN911.22

刘绍华, 李明豪, 李兆歆, 毛天露, 刘京. 一种联合采样的神经网络光场[J]. 北京邮电大学学报, 2021, 44(6): 109-115.

LIU Shao-hua, LI Ming-hao, LI Zhao-xin, MAO Tian-lu, LIU Jing. Neural Radiance Field by Joint Sampling[J]. Journal of Beijing University of Posts and Telecommunications, 2021, 44(6): 109-115.

参考文献

[1] Levoy M, Hanrahan P.Light field rendering[C]//Proceedings of the 23rd Annual Conference on Computer Graphics and Interactive Techniques-SIGGRAPH'96.New York:ACM Press, 1996:31-42.
[2] Wood D N, Azuma D I, Aldinger K, et al.Surface light fields for 3D photography[C]//Proceedings of the 27th Annual Conference on Computer Graphics and Interactive Techniques-SIGGRAPH'00.New York:ACM Press, 2000:287-296.
[3] Lombardi S, Simon T, Saragih J, et al.Neural volumes[J].ACM Transactions on Graphics, 2019, 38(4):1-14.
[4] Mildenhall B, Srinivasan P P, Ortiz-Cayon R, et al.Local light field fusion[J].ACM Transactions on Graphics, 2019, 38(4):1-14.
[5] Liu Lingjie, Gu Jiatao, Lin Kyaw Zaw, et al.Neural sparse voxel fields[C]//Advances in Neural Information Processing Systems 33, NeurlPS 2020.Cambridge:The MIT Press, 2020:15651-15663.
[6] Mildenhall B, Srinivasan P P, Tancik M, et al.NeRF:representing scenes as neural radiance fields for view synthesis[M]//Computer Vision-ECCV 2020.Cham:Springer International Publishing, 2020:405-421.
[7] Lindell D B, Martel J N P, Wetzstein G.AutoInt:automatic integration for fast neural volume rendering[C]//Conference on Computer Vision and Pattern Recognition (CVPR).Piscataway:IEEE Press, 2021:14556-14565.
[8] Reiser C, Peng Songyou, Liao Yiyi, et al.KiloNeRF:speeding up neural radiance fields with thousands of tiny MLPs[EB/OL].(2021-03-25)[2021-04-30].https://arxiv.org/abs/2103.13744.
[9] Landy M, Movshon J.The plenoptic function and the elements of early vision[C]//Computational Models of Visual Processing.Cambridge:MIT Press, 1991:3-20.
[10] Rahaman N, Baratin A, Arpit D, et al.On the spectral bias of neural networks[C]//International Conference on Machine Learning.Cambridge:PMLR, 2019:5301-5310.
[11] Sitzmann V, Zollh fer M, Wetzstein G.Scene representation networks:continuous 3D-structure-aware neural scene representations[M].Red Hook:Curran Associates Inc, 2019:1121-1132.
[12] Sch nberger J L, Frahm J M.Structure-from-motion revisited[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Las Vegas:IEEE Press, 2016:4104-4113.
[13] Sch nberger J L, Zheng Enliang, Frahm J M, et al.Pixelwise view selection for unstructured multi-view stereo[C]//Computer Vision-ECCV 2016.Cham:Springer International Publishing, 2016:501-518.
[14] Kingma D, Ba J.Adam:a method for stochastic optimization[EB/OL].(2014-12-22)[2021-04-30].https://arxiv.org/abs/1412.6980v8.

[1]	袁建国余林峰游薇庞宇. 一种优化错误模式集的极化码 SCL-Flip 译码算法[J]. 北京邮电大学学报, 2023, 46(4): 76-82.
[2]	田心记王坤. IRS辅助下行NOMA系统中最大化折中的方法[J]. 北京邮电大学学报, 0, (): 0-0.
[3]	黄龙杨张楠刘笑笑申滨. 异构蜂窝网络中基于雾节点协作贡献度的计算卸载算法[J]. 北京邮电大学学报, 2023, 46(2): 37-42.
[4]	张浩冯春燕杨佳汇郭彩丽周博文. 面向语义通信的3D骨骼点数据编码与压缩方法[J]. 北京邮电大学学报, 2022, 45(6): 62-69.
[5]	王则予张梦菲孙耀华彭木根. 星上透明转发非地面网络中的切换机制研究[J]. 北京邮电大学学报, 2022, 45(6): 105-112.
[6]	朱剑锋孙耀华彭木根. 低轨卫星通信系统的前导序列设计研究[J]. 北京邮电大学学报, 2022, 45(6): 78-84.
[7]	牛凯吴泊霖戴金晟王森袁弋非. 面向6G系统的极化编码调制[J]. 北京邮电大学学报, 2022, 45(6): 1-11.
[8]	张顺外夏子寒. 基于SWIPT的乘积Polar编码协作构造与分析[J]. 北京邮电大学学报, 2022, 45(5): 54-59.
[9]	赵海英, 解光鹏, 高子惠. 基于数学规则的龟背纹自动生成方法[J]. 北京邮电大学学报, 2022, 45(4): 110-115.
[10]	赵海英解光鹏高子惠. 基于数学规则的龟背纹自动生成方法[J]. 北京邮电大学学报, 2022, 45(4): 119-125.
[11]	郭世泽, 吕仁健, 何明枢, 张杰, 俞赛赛. 流谱理论及其在网络防御中的应用[J]. 北京邮电大学学报, 2022, 45(3): 19-25.
[12]	王惠琴, 杨丽荣, 彭清斌, 曹明华, 陈丹. 基于离散小波变换的O-OFDM-DIM[J]. 北京邮电大学学报, 2022, 45(3): 57-63.
[13]	赵海英, 李焕洪, 侯小刚. 基于感知驱动的纹饰矢量图评价方法[J]. 北京邮电大学学报, 2021, 44(6): 53-58.
[14]	刘小明, 齐涛, 甘露. 一种宽带反射型极化旋转器[J]. 北京邮电大学学报, 2021, 44(6): 20-25,32.
[15]	卓新建, 王文璇, 李慧嘉. 有标记网络中的病毒传播机理及动态特性[J]. 北京邮电大学学报, 2021, 44(3): 73-78.

一种联合采样的神经网络光场

Neural Radiance Field by Joint Sampling

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价