mlsanityData to 7 Oct 2026

The Role of Initialization in 3D Gaussian Splatting

arXiv 2603.20714 ↗First posted 21 Mar 2026cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

3D Gaussian Splatting (3DGS) has become the method of choice for photo-realistic novel view synthesis (NVS), due to its efficiency and compelling visual quality. 3DGS represents the scene as a set of 3D Gaussians, parameterized by their position, spatial extent, and view-dependent color. Starting from an initial point cloud, 3DGS refines the Gaussians' parameters to reconstruct a set of training images as accurately as possible. Typically, a sparse Structure-from-Motion point cloud is used as initialization. Thus, in order to obtain a full scene representation, 3DGS methods rely on a densification stage. In this paper, we systematically study how initialization affects 3DGS NVS performance and geometric quality, using several densification strategies. We show that dense initialization does not lead to consistent visual improvements when paired with strong densification. Despite that, it helps in generalization to off-trajectory views and significantly improves geometric accuracy of the scenes. Our code is available at https://github.com/deivse/ivd_splat.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 49 of its 68 references that are in the map. First posted on arXiv 21 Mar 2026.

Builds on

49 of 68 references.

References · 49

  1. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  2. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,764
  3. 20233D Gaussian Splatting for Real-Time Radiance Field Renderingcited by 6,141
  4. 2021Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fieldscited by 2,013
  5. 2019Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfercited by 1,484
  6. 2025VGGT: Visual Geometry Grounded Transformercited by 1,475
  7. 20234D Gaussian Splatting for Real-Time Dynamic Scene Renderingcited by 1,144
  8. 20242D Gaussian Splatting for Geometrically Accurate Radiance Fieldscited by 1,061
  9. 2023Vision Transformers Need Registerscited by 844
  10. 2023Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene Reconstructioncited by 753
  11. 2023ScanNet++: A High-Fidelity Dataset of 3D Indoor Scenescited by 747
  12. 2023Scaffold-GS: Structured 3D Gaussians for View-Adaptive Renderingcited by 684
37 more
  1. 2023pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstructioncited by 668
  2. 2023Mip-Splatting: Alias-free 3D Gaussian Splattingcited by 661
  3. 2025Depth Anything 3: Recovering the Visual Space from Any Viewscited by 626
  4. 2023SuGaR: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction and High-Quality Mesh Renderingcited by 611
  5. 2024MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Imagescited by 553
  6. 2023DL3DV-10K: A Large-Scale Scene Dataset for Deep Learning-based 3D Visioncited by 514
  7. 2024Metric3Dv2: A Versatile Monocular Geometric Foundation Model for Zero-shot Metric Depth and Surface Normal Estimationcited by 389
  8. 2023RoMa: Robust Dense Feature Matchingcited by 296
  9. 2024PGSR: Planar-based Gaussian Splatting for Efficient and High-Fidelity Surface Reconstructioncited by 280
  10. 2024gsplat: An Open-Source Library for Gaussian Splattingcited by 279
  11. 2024Gaussian Opacity Fields: Efficient Adaptive Surface Reconstruction in Unbounded Scenescited by 226
  12. 2023GS-IR: 3D Gaussian Splatting for Inverse Renderingcited by 220
  13. 20243D Gaussian Splatting as Markov Chain Monte Carlocited by 210
  14. 2024Mini-Splatting: Representing Scenes with a Constrained Number of Gaussianscited by 208
  15. 2024WildGaussians: 3D Gaussian Splatting in the Wildcited by 150
  16. 2023Relightable 3D Gaussians: Realistic Point Cloud Relighting with BRDF Decomposition and Ray Tracingcited by 138
  17. 2023Multi-Scale 3D Gaussian Splatting for Anti-Aliased Renderingcited by 137
  18. 2024Revising Densification in Gaussian Splattingcited by 127
  19. 2024StopThePop: Sorted Gaussian Splatting for View-Consistent Real-time Renderingcited by 107
  20. 2024AbsGS: Recovering Fine Details for 3D Gaussian Splattingcited by 80
  21. 2024InstantSplat: Sparse-view Gaussian Splatting in Secondscited by 67
  22. 2024Analytic-Splatting: Anti-Aliased 3D Gaussian Splatting via Analytic Integrationcited by 66
  23. 2024GS^3: Efficient Relighting with Triple Gaussian Splattingcited by 42
  24. 2024Relaxing Accurate Initialization Constraint for 3D Gaussian Splattingcited by 33
  25. 2024MVPGS: Excavating Multi-view Priors for Gaussian Splatting from Sparse Input Viewscited by 29
  26. 2025Sparse2DGS: Geometry-Prioritized Gaussian Splatting for Surface Reconstruction from Sparse Viewscited by 25
  27. 2024NerfBaselines: Consistent and Reproducible Evaluation of Novel View Synthesis Methodscited by 24
  28. 2025EDGS: Eliminating Densification for Efficient Convergence of 3DGScited by 21
  29. 2025Sharp Monocular View Synthesis in Less Than a Secondcited by 19
  30. 2025NeuralGS: Bridging Neural Fields and 3D Gaussian Splatting for Compact 3D Representationscited by 13
  31. 2024Evaluating Alternatives to SFM Point Cloud Initialization for Gaussian Splattingcited by 9
  32. 2025VA-GS: Enhancing the Geometric Representation of Gaussian Splatting via View Alignmentcited by 6
  33. 2025BOGausS: Better Optimized Gaussian Splattingcited by 5
  34. 2025Improving Densification in 3D Gaussian Splatting for High-Fidelity Renderingcited by 5
  35. 2025GDGS: 3D Gaussian Splatting Via Geometry-Guided Initialization And Dynamic Density Controlcited by 2
  36. 2025HBSplat: Robust Sparse-View Gaussian Reconstruction with Hybrid-Loss Guided Depth and Bidirectional Warpingcited by 1
  37. 2025Initialize to Generalize: A Stronger Initialization Pipeline for Sparse-View 3DGScited by 1

Built on by

Not cited yet.