Depth-Gradient-Guided Decoupled Optimization in 3D Gaussian Splatting for Sparse-View Reconstruction
Source abstract
Abstract
3D Gaussian Splatting (3DGS) achieves high-quality novel-view synthesis under dense inputs but suffers from severe overfitting and geometric artifacts in sparse-view settings. Existing approaches introduce monocular depth priors to alleviate this issue, yet they typically apply uniform geometric constraints across the entire scene, ignoring structural heterogeneity between flat and edge regions, which leads to a trade-off between global smoothness and local detail preservation. To address this limitation, we propose Depth-Gradient-Guided Decoupled Optimization for Gaussian Splatting (DGDO-GS), a geometry-aware framework designed for sparse-view reconstruction. Specifically, depth gradients are exploited to capture structural variations and partition the scene into flat and edge regions. Global consistency constraints are imposed on flat areas to enhance stability, while local sharpening regularization is applied to edge regions to preserve fine-grained details. An orthogonal decoupling strategy further separates position and shape optimization, enabling accurate geometric recovery under monocular depth supervision while maintaining high-frequency texture fidelity. Experiments on LLFF, Mip-NeRF360, and Blender demonstrate that DGDO-GS consistently outperforms existing methods in reconstruction quality.
Source: crossref · Quality: baseline
Research context
OpenAlex topics: 3D Shape Modeling and Analysis · Advanced Vision and Imaging · Computer Graphics and Visualization Techniques
SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.
Affiliations
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
Key Laboratory of Target Cognition and Application Technology, Chinese Academy of Sciences, Beijing 100190, China
School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China
References (41)
Show source-provided references
1. Avidan, S., and Shashua, A. (1997, January 17–19). Novel view synthesis in tensor space. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, San Juan, PR, USA.
2. NeRF: Representing scenes as neural radiance fields for view synthesis10.1145/3503250 ↗
3. MLP-Mixer: An all-MLP architecture for vision
4. Cui, Z., Zhang, Y., Wang, X., and Li, G. (2024, January 20–27). Aleth-NeRF: Illumination adaptive NeRF with concealing field assumption. Proceedings of the AAAI Conference on Artificial Intelligence, Vancouver, BC, Canada.10.1609/aaai.v38i2.27908 ↗
5. Lee, J.L., Li, C., and Lee, G.H. (2024, January 16–22). DISR-NeRF: Diffusion-guided view-consistent super-resolution NeRF. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA.10.1109/CVPR52733.2024.01943 ↗
6. Li, H., Wang, P., Liu, Y., and Li, G. (2024, January 16–22). GP-NeRF: Generalized perception NeRF for context-aware 3D scene understanding. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA.10.1109/CVPR52733.2024.02051 ↗
7. 3D Gaussian splatting for real-time radiance field rendering10.1145/3592433 ↗
8. Zhu, Z., Chen, X., Xu, M., Wang, C., Li, H., and Lin, S. (October, January 29). FSGS: Real-time few-shot view synthesis using Gaussian splatting. Proceedings of the European Conference on Computer Vision, Milan, Italy.
9. Li, J., Zhang, Y., and Li, G. (2024, January 16–22). DNGaussian: Optimizing sparse-view 3D Gaussian radiance fields with global-local depth normalization. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA.10.1109/CVPR52733.2024.01963 ↗
10. Local light field fusion: Practical view synthesis with prescriptive sampling guidelines10.1145/3306346.3322980 ↗
11. Jensen, R., Dahl, A., Vogiatzis, G., Nistér, D., and Stewénius, H. (2014, January 23–28). Large scale multi-view stereopsis evaluation. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA.10.1109/CVPR.2014.59 ↗
12. Barron, J.T., Mildenhall, B., Tancik, M., Srinivasan, P.P., Funkhouser, T., and Ng, R. (2022, January 18–24). Mip-NeRF 360: Unbounded anti-aliased neural radiance fields. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, New Orleans, LA, USA.10.1109/CVPR52688.2022.00539 ↗
13. Dai, S., and Wang, S. (2025). HR-NeRF: Advancing realism and accuracy in highlight scene representation. Front. Neurorobot., 19.10.3389/fnbot.2025.1558948 ↗
14. Nguyen, H., Li, B.R., and Nguyen, T. (2025). DWTNeRF: Boosting few-shot neural radiance fields via discrete wavelet transform. arXiv.
15. Chen, A., Xu, Z., Zhao, F., Zhang, X., Xiang, F., Yu, J., and Su, H. (2022). Tensorf: Tensorial radiance fields. Proceedings of the European Conference on Computer Vision, Tel Aviv, Israel, 23–27 October 2022, Springer.10.1007/978-3-031-19824-3_20 ↗
16. Fridovich-Keil, S., Yu, A., Tancik, M., and Ng, R. (2022, January 18–24). Plenoxels: Radiance fields without neural networks. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, New Orleans, LA, USA.10.1109/CVPR52688.2022.00542 ↗
17. TFS-NeRF: Template-free NeRF for semantic 3D reconstruction of dynamic scene
18. NeRF-FF: A plug-in method to mitigate defocus blur for runtime optimized neural radiance fields10.1007/s00371-024-03507-y ↗
19. Luiten, J., Kopanas, G., Leibe, B., and Ramanan, D. (2024). Dynamic 3D Gaussians: Tracking by persistent dynamic view synthesis. Proceedings of the International Conference on 3D Vision (3DV), Davos, Switzerland, 18–21 March 2024, IEEE.10.1109/3DV62453.2024.00044 ↗
20. Tang, J., Ren, J., Zhou, H., Liu, Z., and Zeng, G. (2023). DreamGaussian: Generative Gaussian splatting for efficient 3D content creation. arXiv.
21. Wu, G., Yi, T., Fang, J., Xie, L., Zhang, X., Wei, W., Liu, W., Tian, Q., and Wang, X. (2024, January 16–22). 4D Gaussian splatting for real-time dynamic scene rendering. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA.10.1109/CVPR52733.2024.01920 ↗
22. TVNeRF: Improving few-view neural volume rendering with total variation maximization10.1016/j.knosys.2024.112273 ↗
23. Xu, J., Deng, K., Fan, Z., Wang, S., Xie, J., and Yang, J. (2025, January 19–20). AD-GS: Object-Aware B-Spline Gaussian Splatting for Self-Supervised Autonomous Driving. Proceedings of the IEEE/CVF International Conference on Computer Vision, Honolulu, HI, USA.
24. TD-NeRF: Transfer learning and diffusion regulation based NeRF for scene perception
25. Uy, M.A., Martin-Brualla, R., Guibas, L., and Li, K. (2023, January 17–24). Scade: NeRFs from space carving with ambiguity-aware depth estimates. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, BC, Canada.10.1109/CVPR52729.2023.01585 ↗
26. Kong, H., Yang, X., and Wang, X. (2025, January 10–17). Generative Sparse-View Gaussian Splatting. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Nashville, TN, USA.10.1109/CVPR52734.2025.02491 ↗
27. Du, S., Liu, J., Chen, Q., Chen, H.X., Mu, T.J., and Yang, S. (2025). RGE-GS: Reward-guided expansive driving scene reconstruction via diffusion priors. arXiv.
28. 3DGS-Enhancer: Enhancing unbounded 3D Gaussian splatting with view-consistent 2D diffusion priors
29. Kim, M., Jeon, S., Cho, I., Yoo, M., and Kim, S.J. (2025, January 19–20). ExploreGS: Explorable 3D Scene Reconstruction with Virtual Camera Samplings and Diffusion Priors. Proceedings of the IEEE/CVF International Conference on Computer Vision, Honolulu, HI, USA.
30. Paliwal, A., Zhou, X., Ye, W., Xiong, J., Ranjan, R., and Kalantari, N.K. (2025). RI3D: Few-shot Gaussian splatting with repair and inpainting diffusion priors. arXiv.
31. Cheng, K., Long, X., Yang, K., Yao, Y., Yin, W., Ma, Y., Wang, W., and Chen, X. (2024, January 21–27). GaussianPro: 3D Gaussian splatting with progressive propagation. Proceedings of the International Conference on Machine Learning (ICML), Vienna, Austria.
32. Schönberger, J.L., and Frahm, J.-M. (2016, January 27–30). Structure-from-Motion Revisited. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA.10.1109/CVPR.2016.445 ↗
33. Schönberger, J.L., Zheng, E., Pollefeys, M., and Frahm, J.-M. (2016, January 11–14). Pixelwise View Selection for Unstructured Multi-View Stereo. Proceedings of the European Conference on Computer Vision (ECCV), Amsterdam, The Netherlands.10.1007/978-3-319-46487-9_31 ↗
34. Ranftl, R., Bochkovskiy, A., and Koltun, V. (2021, January 10–17). Vision Transformers for Dense Prediction. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Montreal, QC, Canada.10.1109/ICCV48922.2021.01196 ↗
35. FeatureGS: Eigenvalue-feature optimization in 3D Gaussian splatting for geometrically accurate and artifact-reduced reconstruction10.1016/j.ophoto.2025.100100 ↗
36. RegNeRF: Regularizing Neural Radiance Fields for View Synthesis from Sparse Inputs
37. Schwarz, K., Liao, Y., Niemeyer, M., and Geiger, A. (2021, January 10–17). DietNeRF: Distilling Neural Radiance Fields for Few-Shot View Synthesis. Proceedings of the IEEE/CVF International Conference on Computer Vision, Montreal, QC, Canada.
38. Chen, X., Zhang, Y., and Li, G. (2023, January 17–24). FreeNeRF: Frequency Modulation for Few-Shot Neural Radiance Fields. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, BC, Canada.
39. Zhang, H., Wang, Y., and Liu, Z. (2022, January 23–27). SparseNeRF: Depth Ranking Guided Neural Radiance Fields for Sparse View Synthesis. Proceedings of the European Conference on Computer Vision, Tel Aviv, Israel.
40. Zhang, R., Isola, P., Efros, A.A., Shechtman, E., and Wang, O. (2018, January 18–23). The Unreasonable Effectiveness of Deep Features as a Perceptual Metric. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA.10.1109/CVPR.2018.00068 ↗
41. PyTorch: An Imperative Style, High-Performance Deep Learning Library