InfNeRF: Towards Infinite Scale NeRF Rendering with O(log n) Space Complexity DOI Open Access
Jiabin Liang, Lanqing Zhang, Zhuoran Zhao

и другие.

Опубликована: Дек. 3, 2024

The conventional mesh-based Level of Detail (LoD) technique, exemplified by applications such as Google Earth and many game engines, exhibits the capability to holistically represent a large scene even Earth, achieves rendering with space complexity \(\mathcal {O}(\log n)\). This constrained data requirement not only enhances efficiency but also facilitates dynamic fetching, thereby enabling seamless 3D navigation experience for users.In this work, we extend proven LoD technique Neural Radiance Fields (NeRF) introducing an octree structure scenes in different scales. innovative approach provides mathematically simple elegant representation n)\), aligned techniques. We present novel training strategy that maintains {O}(n)\). allows parallel minimal overhead, ensuring scalability our proposed method. Our contribution is extending capabilities existing techniques establishing foundation scalable efficient large-scale using NeRF structures. Code checkpoints are available at: https://jiabinliang.github.io/InfNeRF.io/

Язык: Английский

Scene reconstruction techniques for autonomous driving: a review of 3D Gaussian splatting DOI Creative Commons

Huixin Zhu,

Zhili Zhang, Junyang Zhao

и другие.

Artificial Intelligence Review, Год журнала: 2024, Номер 58(1)

Опубликована: Ноя. 30, 2024

As the latest research result of explicit radiated field technology, 3D Gaussian Splatting (3D GS) replaces implicit expression represented by Neural Radiated Field (NeRF) and has become hottest direction in scene reconstruction. Given innovative work vigorous development GS autonomous driving, this paper comprehensively reviews summarizes existing related to showcase evolution technology possible future directions. First, overall background is introduced based on two aspects reconstruction methods progress. Second, relevant knowledge points core formulas clarify mathematical mechanism are presented. Third, primary applications automatic driving presented through new perspective synthesis, understanding, simultaneous localization map building (SLAM). Finally, frontier directions described, including structure optimization, 4D reconstruction, cross-domain research. This may provide an effective convenient pathway for researchers understand, explore, apply novel method, promote application driving.

Язык: Английский

Процитировано

3
High-Fold 3D Gaussian Splatting Model Pruning Method Assisted by Opacity DOI Creative Commons

Shiyu Qiu,

Chunlei Wu, Zhe Wan

и другие.

Applied Sciences, Год журнала: 2025, Номер 15(3), С. 1535 - 1535

Опубликована: Фев. 3, 2025

Recent advancements in 3D scene representation have underscored the potential of Neural Radiance Fields (NeRFs) for producing high-fidelity renderings complex scenes. However, NeRFs are hindered by significant computational burden volumetric rendering. To address this, Gaussian Splatting (3DGS) has emerged as an efficient alternative, utilizing Gaussian-based representations and rasterization techniques to achieve faster rendering speeds without sacrificing image quality. Despite these advantages, large number points associated internal parameters result high storage demands. this challenge, we propose a pruning strategy applied during densification phases. Our approach integrates learnable masks with contribution-based mechanism, further enhanced opacity update facilitate process. This method effectively eliminates redundant those minimal contributions construction. Additionally, parameter compression phase, employ combination teacher–student models vector quantization compress spherical harmonic coefficients. Extensive experimental results demonstrate that our reduces requirements original over 30 times, only minor degradation

Язык: Английский

Процитировано

0
A novel framework utilizing 3D Gaussian Splatting to construct building geometry for urban wind simulations DOI

Peisheng Zhao,

Chao Li, J. S. Jiang

и другие.

Sustainable Cities and Society, Год журнала: 2025, Номер unknown, С. 106237 - 106237

Опубликована: Фев. 1, 2025

Язык: Английский

Процитировано

0
Li-GS: a fast 3D Gaussian reconstruction method assisted by LiDAR point clouds DOI Creative Commons
Wenzhuo Chen, Ruofei Zhong,

Kangfei Wang

и другие.

Big Earth Data, Год журнала: 2025, Номер unknown, С. 1 - 25

Опубликована: Март 21, 2025

Язык: Английский

Процитировано

0
AAGS: Appearance-Aware 3D Gaussian Splatting with Unconstrained Photo Collections DOI
Wei Zhang, Zhiyang Guo, Wengang Zhou

и другие.

Multimedia Systems, Год журнала: 2025, Номер 31(2)

Опубликована: Март 28, 2025

Язык: Английский

Процитировано

0
Depth-Consistent 3d Gaussian Splatting Via Physical Defocus Modeling and Multi-View Geometric Supervision DOI
Yu Deng, Baozhu Zhao, Junyan Su

и другие.

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
On Scaling Up 3D Gaussian Splatting Training DOI

Hexu Zhao,

Haoyang Weng,

Daohan Lu

и другие.

Lecture notes in computer science, Год журнала: 2025, Номер unknown, С. 14 - 36

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
基于三维高斯溅射技术的可微分渲染研究进展 DOI

高建 Gao Jian,

陈林卓 Chen Linzhuo,

沈秋 Shen Qiu

и другие.

Laser & Optoelectronics Progress, Год журнала: 2024, Номер 61(16), С. 1611010 - 1611010

Опубликована: Янв. 1, 2024

Процитировано

0
InfNeRF: Towards Infinite Scale NeRF Rendering with O(log n) Space Complexity DOI Open Access
Jiabin Liang, Lanqing Zhang, Zhuoran Zhao

и другие.

Опубликована: Дек. 3, 2024

The conventional mesh-based Level of Detail (LoD) technique, exemplified by applications such as Google Earth and many game engines, exhibits the capability to holistically represent a large scene even Earth, achieves rendering with space complexity \(\mathcal {O}(\log n)\). This constrained data requirement not only enhances efficiency but also facilitates dynamic fetching, thereby enabling seamless 3D navigation experience for users.In this work, we extend proven LoD technique Neural Radiance Fields (NeRF) introducing an octree structure scenes in different scales. innovative approach provides mathematically simple elegant representation n)\), aligned techniques. We present novel training strategy that maintains {O}(n)\). allows parallel minimal overhead, ensuring scalability our proposed method. Our contribution is extending capabilities existing techniques establishing foundation scalable efficient large-scale using NeRF structures. Code checkpoints are available at: https://jiabinliang.github.io/InfNeRF.io/

Язык: Английский

Процитировано

0