A Review of Neural Radiance Fields and 3D Gaussian Splatting for 3D Reconstruction
DOI:
https://doi.org/10.61173/b5v3xb72Keywords:
Neural Radiance Fields, 3D Gaussian Splat-ting method, Representation Paradigm, Frames Per Sec-ondAbstract
Within the disciplines of computer vision and deep learning, the ability to construct 3D models from data has become a fundamental capability. A recent wave of progress has significantly advanced the two leading methods for scene representation:Neural Radiance Fields for implicit modeling and 3D Gaussian Splatting for explicit construction. This review aims to build a systematic cognitive framework of 3D reconstruction technology for readers by comparing the performance of these two technologies and their variants in static and dynamic scenes. This review selects representative evaluation parameters such as Peak Signal-to-Noise Ratio. By collecting experimental data from public datasets, it compares, organizes and summarizes the high-quality rendering ability of neural radiance field technology for geometric details and the high-speed real-time rendering ability of 3D Gaussian splatting. Looking ahead, this paper proposes key development directions such as the integration of expression paradigms and overcoming the slow speed of implicit expression, hoping to provide a structural knowledge framework and research inspiration for the professional field.
References
[1] Wang P, Liu Y, Liu Z, et al. NeRF-based 3D human-body reconstruction: a survey. arXiv preprint arXiv:2305.16823, 2023.
[2] Mildenhall B, Srinivasan P P, Tancik M, et al. NeRF: representing scenes as neural radiance fields for view synthesis. European Conference on Computer Vision (ECCV), 2020: 405- 421.
[3] Kerbl B, Kopanas G, Leimkühler T, Drettakis G. 3D gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics, 2023, 42(4): 1-14.
[4] Mildenhall B, Srinivasan P P, Tancik M, Barron J T, Ramamoorthi R, Ng R. NeRF: representing scenes as neural radiance fields for view synthesis. European Conference on Computer Vision (ECCV), 2020: 405-421.
[5] Yu A, Li R M, Tancik M, et al. PlenOctrees for real-time rendering of neural radiance fields. IEEE/CVF International Conference on Computer Vision (ICCV), 2021: 5752-5761.
[6] Barron J T, Mildenhall B, Verbin D, et al. Zip-NeRF: anti-aliased grid-based neural radiance fields. IEEE/CVF International Conference on Computer Vision (ICCV), 2023: 19755-19764.
[7] Müller T, Evans A, Schied C, Keller A. Instant neural graphics primitives with a multiresolution hash encoding. ACM Transactions on Graphics, 2022, 41(4): 102.
[8] Tretschk E, Tewari A, Golyanik V, et al. Non-rigid neural radiance fields: reconstruction and novel view synthesis of a dynamic scene from monocular video. IEEE/CVF International Conference on Computer Vision (ICCV), 2021: 12959-12970.
[9] Park K, Sinha U, Hedman P, et al. HyperNeRF: a higherdimensional representation for topologically varying neural radiance fields. ACM Transactions on Graphics, 2021, 40(6): 238.
[10] Xie X, Guo W, Li J, Liu J, Xu J. Deform2NeRF: non-rigid deformation and 2D-3D feature fusion with cross-attention for dynamic human reconstruction. Electronics, 2023, 12(21): 4382
[11] Kerbl B, Kopanas G, Leimkühler T, Drettakis G. 3D gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics, 2023, 42(4): 139.
[12] Liu Y, Guan H, Luo C, et al. CityGaussian: real-time highquality large-scale scene rendering with gaussians. European Conference on Computer Vision (ECCV), 2024.
[13] Chen Y, Fan L, Wang Y, Liu X, Wang J. Split-SSG: splitting screen-space-accurate 3D gaussian splatting for high-fidelity real-time rendering. arXiv preprint arXiv:2312.01317, 2023
[14] Lee J C, Rho D, Sun X, Ko J H, Park E. Compact 3D gaussian representation for radiance field. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024.
[15] Wu G, et al. 4D gaussian splatting for real-time dynamic scene rendering. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024: 20310-20320.
[16] Lin Y, Dai Z, Zhu S, Yao Y. Gaussian-Flow: 4D reconstruction with dynamic 3D gaussian particle. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024
[17] Zheng C, Xue L, Zarate J, Song J. GauSTAR: gaussian surface tracking and reconstruction. arXiv preprint arXiv:2501.10283, 2025.
[18] Wang Z, Bovik A C, Sheikh H R, Simoncelli E P. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing, 2004, 13(4): 600-612.
[19] Yariv L, Hedman P, Reiser C, et al. BakedSDF: Meshing Neural SDFs for Real-Time View Synthesis. ACM Transactions on Graphics (TOG), 2023, 42(4): 1-16.
[20] Ren Z, Chen W, Zhang J, et al. K-Buffers: A Plug-in Method for Enhancing Neural Fields with Multiple Buffers. arXiv preprint arXiv:2405.18318, 2024.
[21] Guédon A, Lepetit V. SplatFlow: Splatting Gaussian Kernels for Dynamic Scenes from Monocular Videos. arXiv preprint arXiv:2311.12337, 2023.
[22] Chen A, Xu Z, Zhao F, et al. MVSNeRF: Fast Generalizable Radiance Field Reconstruction from Multi-View Stereo. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021: 14124-14133.
[23] Miao Z, Yang S, Yu K, et al. EVO-Splat: Evolving 3D Gaussian Splatting for Real-time Dynamic Scene Reconstruction. arXiv preprint arXiv:2403.18946, 2024.
Downloads
Published
Issue
Section
License
Copyright (c) 2025 by the authors.
This work is licensed under a Creative Commons Attribution 4.0 International License.
