Research map / 3D vision / Gaussian splatting
T$^2$-Splat: Adaptive Topology Mesh Splatting with Texture Residuals
Zhihao Tang, Youjia Zhang, Mingbo Zhao, Wei Yang
Read the paperCheck your idea against 137,217 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.
Proposes T^2-Splat, a mesh splatting framework that adaptively adjusts topology and adds texture residuals to improve novel view synthesis on polygonal primitives.
Problem
Naively applying unstructured splatting to triangle/polygonal primitives fails to exploit their topological advantages, causing over-tessellation and degraded high-frequency appearance in novel view synthesis.
Approach
A mesh splatting framework with adaptive topology and texture residuals, so that mesh structure adapts to scene content and high-frequency detail is carried by residual textures rather than extra geometry.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Safe RDT initialization (ablation) | MeshSplatting baseline · Mip-NeRF 360 (9 scenes) | faces: 5.3M -> 3.0M | Improved |
| add Safe RDT initialization (ablation) | baseline · Mip-NeRF 360 | LPIPS: 0.30 -> 0.31 | No clear effect |
| add Dynamic Topology Evolution (edge collapse + vertex split) (ablation) | Safe RDT mesh · Mip-NeRF 360 | PSNR (dB): 24.81 -> 25.12 | Improved |
| add DTE on top of 8x8 textures (ablation) | Safe RDT + MipTex · Mip-NeRF 360 | PSNR (dB): 25.31 -> 25.49 | Improved |
| add 4x4 residual Mip-textures (ablation) | DTE mesh · Mip-NeRF 360 | PSNR (dB): 25.12 -> 25.43 | Improved |
| increase texture resolution from 4x4 to 8x8 (ablation) | T2-Splat · Mip-NeRF 360 | LPIPS: 0.26 -> 0.23 | Improved |
| add DTE (cost: training time) (ablation) | Safe RDT · Mip-NeRF 360 | training time (min): 43 -> 57 | Hurt |
| reduce face budget to 50% (ablation) | T2-Splat · Garden | PSNR (dB): 25.68 -> 25.38 | No clear effect |
| reduce face budget to 10% (ablation) | T2-Splat · Garden | PSNR (dB): 25.68 -> 23.93 | Hurt |
| T2-Splat vs MeshSplatting on Tanks & Temples | T2-Splat · Tanks & Temples | PSNR (dB): 21.12 vs 20.52 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MeshSplatting — MeshSplatting: Differentiable Rendering with Opaque Meshes (CVPR 2026) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Experimental setup
Builds on: triangle splatting
Closest papers by meaning
- MeshSplatting: Differentiable Rendering with Opaque Meshes (CVPR 2026 · similarity 0.90)
- Tile-wise vs. Image-wise: Random-Tile Loss and Training Paradigm for Gaussian Splatting (ICCV 2025 · similarity 0.88)
- UniMGS: Unifying Mesh and 3D Gaussian Splatting with Single-Pass Rasterization and Proxy-Based Deformation (AAAI 2026 · similarity 0.88)
- Topology-Aware 3D Gaussian Splatting: Leveraging Persistent Homology for Optimized Structural Integrity (AAAI 2025 · similarity 0.88)
- Swift4D: Adaptive divide-and-conquer Gaussian Splatting for compact and efficient reconstruction of dynamic scene (ICLR 2025 · similarity 0.88)
- TetSphere Splatting: Representing High-Quality Geometry with Lagrangian Volumetric Meshes (ICLR 2025 · similarity 0.87)
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.
