ANeRF: Adversarial neural radiance fields for efficient 3D-aware synthesis

Yinghao Peng; Fangqing Gu; Lei Chen; Yinghao Peng; Fangqing Gu; Lei Chen

doi:10.3934/era.2026229

Electronic Research Archive

2026, Volume 34, Issue 8: 5166-5183. doi: 10.3934/era.2026229

Previous Article Next Article

Research article Special Issues

ANeRF: Adversarial neural radiance fields for efficient 3D-aware synthesis

School of Mathematics and Statistics, Guangdong University of Technology, Guangdong, China

Received: 03 April 2026 Revised: 19 May 2026 Accepted: 05 June 2026 Published: 16 June 2026

Conventional 3D image reconstruction algorithms suffer from fundamental limitations in disentangling 3D scene geometry from rendering processes, which results in poor generalization to novel viewpoints and object poses. To address this, we propose adversarial neural radiance fields (NeRF) for efficient 3D-aware synthesis. By reformulating NeRF as the generator within an adversarial paradigm, our approach bypasses traditional encoding-decoding pipelines through a direct coordinate-to-pixel transformation, thus significantly reducing the computational overhead. This bidirectional enhancement enables a generative adversarial network (GAN) to leverage NeRF's efficient 3D-aware rendering, while NeRF benefits from adversarial regularization to overcome scene-specific overfitting. Extensive evaluations demonstrate that the proposed adversarial NeRF (ANeRF) performs well on benchmark datasets. ANeRF achieves a 3 dB improvement over traditional NeRF in terms of the peak signal-to-noise ratio (PSNR) upon convergence, while accelerating convergence to $ 6 \times 10^{-3} $ loss in 34.4% fewer iterations. The framework maintains exceptional efficiency with 31.42 minutes per epoch and a memory footprint of 5.25 GB. It represents a 43% memory reduction compared to the baseline methods. This explicit 3D representation enables robust multiview consistency and precise scene manipulation, thus establishing a high-efficiency architectural foundation for potential Augmented Reality/Virtual Reality (AR/VR) content generation and digital twin simulations.
- neural radiance field,
- generative adversarial networks,
- adversarial training,
- 3D-aware image synthesis,
- coordinate-based generative modeling
Citation: Yinghao Peng, Fangqing Gu, Lei Chen. ANeRF: Adversarial neural radiance fields for efficient 3D-aware synthesis[J]. Electronic Research Archive, 2026, 34(8): 5166-5183. doi: 10.3934/era.2026229

Related Papers:

Abstract

Conventional 3D image reconstruction algorithms suffer from fundamental limitations in disentangling 3D scene geometry from rendering processes, which results in poor generalization to novel viewpoints and object poses. To address this, we propose adversarial neural radiance fields (NeRF) for efficient 3D-aware synthesis. By reformulating NeRF as the generator within an adversarial paradigm, our approach bypasses traditional encoding-decoding pipelines through a direct coordinate-to-pixel transformation, thus significantly reducing the computational overhead. This bidirectional enhancement enables a generative adversarial network (GAN) to leverage NeRF's efficient 3D-aware rendering, while NeRF benefits from adversarial regularization to overcome scene-specific overfitting. Extensive evaluations demonstrate that the proposed adversarial NeRF (ANeRF) performs well on benchmark datasets. ANeRF achieves a 3 dB improvement over traditional NeRF in terms of the peak signal-to-noise ratio (PSNR) upon convergence, while accelerating convergence to $ 6 \times 10^{-3} $ loss in 34.4% fewer iterations. The framework maintains exceptional efficiency with 31.42 minutes per epoch and a memory footprint of 5.25 GB. It represents a 43% memory reduction compared to the baseline methods. This explicit 3D representation enables robust multiview consistency and precise scene manipulation, thus establishing a high-efficiency architectural foundation for potential Augmented Reality/Virtual Reality (AR/VR) content generation and digital twin simulations.

References

[1]	A. Chatziagapi, G. G. Chrysos, D. Samaras, MI-NeRF: Learning a single NeRF for multiple identities, in European Conference on Computer Vision, (2025), 451–469.
[2]	S. Basak, P. Corcoran, R. McDonnell, M. Schukat, 3D Face-model reconstruction from a single image: A feature aggregation approach using hierarchical transformer with weak supervision, Neural Networks, 156 (2022), 108–122. https://doi.org/10.1016/j.neunet.2022.09.019 doi: 10.1016/j.neunet.2022.09.019
[3]	H. Wei, W. Han, X. Dong, J. Shen, Towards high-fidelity 3D portrait generation with rich details by cross-view prior-aware diffusion, in Proceedings of the AAAI Conference on Artificial Intelligence, 40 (2026), 10521–10529. https://doi.org/10.1609/aaai.v40i13.38024
[4]	M. Wang, S. Zhao, X. Dong, J. Shen, High-fidelity and high-efficiency talking portrait synthesis with detail-aware neural radiance fields, IEEE Trans. Visual Comput. Graphics, 31 (2024), 6022–6035. https://doi.org/10.1109/TVCG.2024.3488960 doi: 10.1109/TVCG.2024.3488960
[5]	T. Tsesmelis, L. Palmieri, M. Khoroshiltseva, A. Islam, G. Elkin, O. I. Shahar, et al., {R}e-assembling the past: the RePAIR dataset and benchmark for real world 2D and 3D puzzle solving, in Advances in Neural Information Processing Systems, 37 (2024), 30076–30105.
[6]	C. Huang, Y. Hou, W. Ye, D. Huang, X. Huang, B. Lin, et al., NeRF-Det++: Incorporating semantic cues and perspective-aware depth supervision for indoor multi-view 3D detection, IEEE Trans. Image Process., 34 (2025), 2575–2587. https://doi.org/10.1109/TIP.2025.3560240 doi: 10.1109/TIP.2025.3560240
[7]	P. Ticha, A. Sukop, Patient-reported outcomes in bilateral prophylactic mastectomy with breast reconstruction: A narrative review, Breast, 73 (2024), 103602. https://doi.org/10.1016/j.breast.2023.103602 doi: 10.1016/j.breast.2023.103602
[8]	I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, et al., Generative adversarial nets, in Advances in Neural Information Processing Systems, 27 (2014), 2672–2680.
[9]	X. Qin, H. Shi, X. Dong, S. Zhang, A new method based on generative adversarial networks for multivariate time series prediction, Expert Syst., 41 (2024), e13700.
[10]	U. Demirbaga, Advancing anomaly detection in cloud environments with cutting-edge generative AI for expert systems, Expert Syst., 42 (2025), e13722.
[11]	F. Han, Y. Liang, Q. Ling, H. Han, An improved conditional wasserstein GAN with gradient penalty for gene expression profiling data augmentation based on data segmentation and depth feature constraint, IEEE Trans. Comput. Biol. Bioinf., 22 (2025), 1401–1414. https://doi.org/10.1109/TCBBIO.2025.3560097 doi: 10.1109/TCBBIO.2025.3560097
[12]	C. Ledig, L. Theis, F. Huszár, J. Caballero, A. P. Aitken, A. Tejani, et al., Photo-realistic single image super-resolution using a generative adversarial network, in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (2017), 105–114.
[13]	H. Xu, Q. Li, J. Chen, Highlight removal from a single grayscale image using attentive GAN, Appl. Artif. Intell., 36 (2022), 1988441. https://doi.org/10.1080/08839514.2022.2127598 doi: 10.1080/08839514.2022.2127598
[14]	L. Tran, X. Yin, X. Liu, Disentangled representation learning GAN for pose-invariant face recognition, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, (2017), 1415–1424.
[15]	Z. Cai, Z. Xiong, H. Xu, P. Wang, W. Li, Y. Pan, Generative adversarial networks: A survey toward private and secure applications, ACM Comput. Surv., 54 (2021), 1–38. https://doi.org/10.1145/3459992 doi: 10.1145/3459992
[16]	Y. Bai, Y. Zhang, M. Ding, B. Ghanem, SOD-MTGAN: Small object detection via multi-task generative adversarial network, in Proceedings of the European Conference on Computer Vision (ECCV), (2018), 206–221.
[17]	E. Denton, V. Birodkar, Unsupervised learning of disentangled representations from video, in Proceedings of the 31st International Conference on Neural Information Processing Systems, (2017), 4417–4426.
[18]	S. Gu, D. Chen, J. Bao, F. Wen, B. Zhang, D. Chen, et al., Vector quantized diffusion model for text-to-image synthesis, in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, (2022), 10696–10706.
[19]	J. Lin, Z. Zhou, Q. Qiu, R. Liu, J. Shen, Y. Zhang, Microatoll-GAN: A generative adversarial network for microatoll detection in assisting monitoring Marine Fishery, in 2025 28th International Conference on Computer Supported Cooperative Work in Design (CSCWD), (2025), 189–194.
[20]	K. Li, K. Deb, Q. Zhang, S. Kwong, An evolutionary many-objective optimization algorithm based on dominance and decomposition, IEEE Trans. Evol. Comput., 19 (2015), 694–716.
[21]	X. Feng, Y. He, Y. Wang, C. Wang, Z. Kuang, J. Ding, et al., ZS-SRT: An efficient zero-shot super-resolution training method for neural radiance fields, Neurocomputing, 590 (2024), 127714.
[22]	X. Li, K. Li, G. Zhang, Z. Zhu, P. Wang, Z. Wang, et al., SREGS: Sparse-view gaussian radiance fields with geometric regularization and region exploration, Neural Networks, 191 (2025), 107820. https://doi.org/10.1016/j.neunet.2025.107820 doi: 10.1016/j.neunet.2025.107820
[23]	B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, R. Ng, NeRF: Representing scenes as neural radiance fields for view synthesis, Commun. ACM, 65 (2021), 99–106.
[24]	J. T. Barron, B. Mildenhall, M. Tancik, P. Hedman, R. Martin-Brualla, P. P. Srinivasan, Mip-NeRF: A multiscale representation for anti-aliasing neural radiance fields, in 2021 IEEE/CVF International Conference on Computer Vision (ICCV), (2021), 5835–5844.
[25]	B. Kerbl, G. Kopanas, T. Leimkuehler, G. Drettakis, 3D gaussian splatting for real-time radiance field rendering, ACM Trans. Graphics, 42 (2023), 1–14. https://doi.org/10.1145/3592455 doi: 10.1145/3592455
[26]	S. Fridovich-Keil, A. Yu, M. Tancik, Q. Chen, B. Recht, A. Kanazawa, Plenoxels: Radiance fields without neural networks, in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2022), 5491–5500.
[27]	Q. Xu, Z. Xu, J. Philip, S. Bi, Z. Shu, K. Sunkavalli, et al., Point-NeRF: Point-based neural radiance fields, in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, (2022), 5438–5448.
[28]	T. Müller, A. Evans, C. Schied, A. Keller, Instant neural graphics primitives with a multiresolution hash encoding, ACM Trans. Graphics, 41 (2022), 102:1–102:15. https://doi.org/10.1145/3528223.3530184 doi: 10.1145/3528223.3530184
[29]	J. Liu, H. Cheng, S. Wang, F. Zhao, M. Li, NeRF dynamic scene reconstruction based on motion, semantic information and inpainting, Neurocomputing, 630 (2025), 129653. https://doi.org/10.1016/j.neucom.2025.129653 doi: 10.1016/j.neucom.2025.129653
[30]	R. Mishan, X. Fu, C. Hingu, P. Fajri, Analyzing frequency spectrum and total harmonic distortion for high switching frequency operation of GAN-based filter-less multilevel cascaded H-bridge inverter, e-Prime Adv. Electr. Eng. Electron. Energy, 11 (2025), 100906. https://doi.org/10.1016/j.prime.2025.100906 doi: 10.1016/j.prime.2025.100906
[31]	M. Hassan, F. Forest, O. Fink, M. Mielle, ThermoNeRF: A multi-modal neural radiance field for joint RGB-thermal novel view synthesis of building facades, Adv. Eng. Inf., 65 (2025), 103345. https://doi.org/10.1016/j.aei.2025.103345 doi: 10.1016/j.aei.2025.103345
[32]	I. Petrovska, B. Jutzi, Vision through obstacles—3D geometric reconstruction and evaluation of neural radiance fields (NeRFs), Remote Sens., 16 (2024), 1188. https://doi.org/10.3390/rs16071188 doi: 10.3390/rs16071188
[33]	Y. Zhou, Z. Zeng, A. Chen, X. Zhou, H. Ni, S. Zhang, et al., Evaluating modern approaches in 3D scene reconstruction: NeRF vs gaussian-based methods, in 2024 6th International Conference on Data-driven Optimization of Complex Systems (DOCS), (2024), 926–931.
[34]	X. Wen, K. Sun, T. Chen, Z. Wang, J. She, Q. Zhao, et al., A NeRF-based technique combined depth-guided filtering and view enhanced module for large-scale scene reconstruction, Knowledge-Based Syst., 316 (2025), 113411. https://doi.org/10.1016/j.knosys.2025.113411 doi: 10.1016/j.knosys.2025.113411
[35]	K. Schwarz, Y. Liao, M. Niemeyer, A. Geiger, GRAF: Generative radiance fields for 3D-aware image synthesis, in Advances in Neural Information Processing Systems, 33 (2020), 20154–20166.
[36]	E. R. Chan, M. Monteiro, P. Kellnhofer, J. Wu, G. Wetzstein, pi-GAN: Periodic implicit generative adversarial networks for 3D-aware image synthesis, in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, (2021), 5799–5809.
[37]	M. Aubry, D. Maturana, A. A. Efros, B. C. Russell, J. Sivic, Seeing 3D chairs: Exemplar part-based 2D-3D alignment using a large dataset of CAD models, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, (2014), 3762–3769.
[38]	Z. Liu, P. Luo, X. Wang, X. Tang, Deep learning face attributes in the wild, in Proceedings of the IEEE International Conference on Computer Vision, (2015), 3730–3738.
[39]	X. He, Y. Peng, Fine-grained visual-textual representation learning, IEEE Trans. Circuits Syst. Video Technol., 30 (2019), 520–531. https://doi.org/10.1109/TCSVT.2019.2892802 doi: 10.1109/TCSVT.2019.2892802
[40]	M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, S. Hochreiter, GANs trained by a two time-scale update rule converge to a local nash equilibrium, in Advances in Neural Information Processing Systems, (2017), 6626–6637.
[41]	Y. Sun, J. Fan, MMD Graph kernel: Effective metric learning for graphs via maximum mean discrepancy, in International Conference on Representation Learning (eds. B. Kim, Y. Yue, S. Chaudhuri, K. Fragkiadaki, M. Khan, Y. Sun), (2024), 22274–22307.
[42]	M. G. Martini, Measuring objective image and video quality: On the relationship between SSIM and PSNR for DCT-based compressed images, IEEE Trans. Instrum. Meas., 74 (2025), 1–13. https://doi.org/10.1109/TIM.2025.3529045 doi: 10.1109/TIM.2025.3529045
[43]	A. Gupta, J. Cao, C. Wang, J. Hu, S. Tulyakov, J. Ren, et al., LightSpeed: Light and fast neural light fields on mobile devices, in Advances in Neural Information Processing Systems, 36 (2023), 31021–31037.
[44]	L. Mescheder, A. Geiger, S. Nowozin, Which training methods for GANs do actually converge, in International Conference on Machine Learning, PMLR, (2018), 3481–3490.
[45]	P. Henzler, N. J. Mitra, T. Ritschel, Escaping plato's cave: 3D shape drom adversarial rendering, in Proceedings of the IEEE/CVF International Conference on Computer Vision, (2019), 9984–9993.
[46]	T. Nguyen-Phuoc, C. Li, L. Theis, C. Richardt, Y. L. Yang, HoloGAN: Unsupervised learning of 3D representations from natural images, in Proceedings of the IEEE/CVF International Conference on Computer Vision, (2019), 7588–7597.
[47]	W. Wang, H. L. Liu, K. C. Tan, A surrogate-assisted differential evolution algorithm for high-dimensional expensive optimization problems, IEEE Trans. Cybern., 53 (2023), 2685–2697. https://doi.org/10.1109/TSMC.2023.3281822 doi: 10.1109/TSMC.2023.3281822

Reader Comments

Your name:*

Email:*
© 2026 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)