Research map / 3D vision / Gaussian splatting
Mani-GS: Gaussian Splatting Manipulation with Triangular Mesh
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Mani-GS binds 3D Gaussians to triangles in a local triangle space so that mesh manipulation transfers to 3DGS with self-adaptive rendering, tolerating inaccurate meshes.
Problem
Manipulating 3D Gaussian Splatting scenes freely (large deformation, local edits, soft-body simulation) while preserving photorealistic rendering quality, especially when the proxy mesh is inaccurate.
Approach
Bind Gaussians to each triangle in a per-triangle local coordinate frame and adapt their global position, rotation, and scale from the triangle's transform and edge lengths, so Gaussians can move off the triangle and keep relative geometry after mesh edits.
Key results
- On NeRF Synthetic, Ours with NeuS mesh reaches 33.65 average PSNR (0.966 SSIM, 0.031 LPIPS?) versus SuGaR's 33.52 PSNR and NeRF-Editing's 32.06 PSNR, surpassing SuGaR by at least 1.5 PSNR in every case with NeuS mesh.
- On DTU, Ours achieves 31.50 PSNR, 0.943 SSIM, 0.088 LPIPS, versus SuGaR's 30.06 PSNR and NeuMesh's 28.29 PSNR.
- Binding Gaussians directly on the mesh drops PSNR to 30.87 versus 33.45 for the full method with NeuS mesh.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Mani-GS with a NeuS mesh instead of SuGaR | 3DGS rendering · NeRF Synthetic (8 subjects) | PSNR | Helped |
| Use Mani-GS instead of SuGaR | 3DGS rendering · DTU | PSNR: 30.06 -> 31.50 | Helped |
| Use Mani-GS instead of SuGaR | 3DGS rendering · DTU | LPIPS: 0.136 -> 0.088 | Helped |
| Add triangle shape-aware Gaussian binding and adaptation instead of binding Gaussians directly on mesh (ablation) | Mani-GS · NeRF Synthetic | PSNR: 30.87 -> 33.45 | Helped |
| Add a 3D position offset to mesh-bound Gaussians (ablation) | Mani-GS · NeRF Synthetic | PSNR: 30.87 -> 32.48 | Helped |
Surprising findings
- Adding a fixed position offset to Gaussians improves static fit but produces noise and distortion after manipulation.
Limitations
- Distortions remain when the manipulated region undergoes highly non-rigid deformation.
- Physics simulation on meshes with more than 35K triangles can take hours.
Experimental setup
Datasets: NeRF Synthetic DTU
Benchmarks: NeRF Synthetic DTU
Metrics: PSNR SSIM LPIPS
Baselines: NeRF-Editing SuGaR NeuMesh 3DGS on Mesh Mesh + Offset
Builds on: 3D Gaussian Splatting SuGaR NeuS NeRF-Editing
Closest papers by meaning
- GaussianPro: 3D Gaussian Splatting with Progressive Propagation (ICML 2024 · similarity 0.92)
- SuGaR: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction and High-Quality Mesh Rendering (CVPR 2024 · similarity 0.91)
- 3D-HGS: 3D Half-Gaussian Splatting (CVPR 2025 · similarity 0.91)
- COLMAP-Free 3D Gaussian Splatting (CVPR 2024 · similarity 0.90)
- FlexGS: Train Once, Deploy Everywhere with Many-in-One Flexible 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- GaussianEditor: Swift and Controllable 3D Editing with Gaussian Splatting (CVPR 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.