T2-Splat: Adaptive Topology Mesh Splatting
with Texture Residuals

Zhihao Tang1, Youjia Zhang1, Mingbo Zhao2, Wei Yang1†

1Huazhong University of Science and Technology 2Donghua University

†Corresponding author

NeurIPS 2026

T²-Splat preserves fine text on the Shelf scene. On Garden, its 2.7 million-face mesh balances rendering quality and geometric compactness.
T2-Splat decouples topology adaptation from appearance refinement. Adaptive mesh connectivity allocates geometry where it is needed, while residual per-face Mip-textures preserve fine appearance details on compact meshes.

Video Overview

Adaptive mesh topology and residual per-face textures, from reconstruction to rendering.

Abstract

Triangle-based splatting connects novel view synthesis with conventional graphics pipelines, but often requires excessive tessellation to preserve appearance details. We present T2-Splat, which decouples topology adaptation from appearance refinement. Safe Restricted Delaunay Triangulation (Safe RDT) initializes mesh connectivity. Dynamic Topology Evolution (DTE) then adapts triangle allocation through edge collapse and vertex split operations. Residual per-face Mip-textures then refine appearance on the fixed mesh using scale-adaptive filtering. Experiments on Mip-NeRF 360, Tanks & Temples, and DTU demonstrate competitive results with compact meshes. T2-Splat improves PSNR by 0.71 dB over MeshSplatting on Mip-NeRF 360. Additional compression experiments demonstrate flexible control of mesh complexity, supporting face budgets as low as 10% of the original face count.

Approach

A two-stage pipeline separates geometry and topology optimization from appearance refinement. The first stage constructs a connected mesh and adapts its triangle allocation. The second stage recovers high-frequency details while keeping the geometry fixed.

Stage I constructs connectivity with Safe RDT and adapts topology through edge collapse and vertex split. Stage II jointly optimizes vertex spherical harmonics and residual Mip-textures on the fixed mesh.
Stage I evolves mesh connectivity. Stage II refines appearance with scale-adaptive texture filtering.

Visual Comparisons

Drag the divider in each image to compare the same view. You can also focus a slider and use the arrow keys.

T squared Splat rendering of the Shelf scene
MeshSplatting rendering of the Shelf scene
MeshSplattingT2-Splat
ShelfSTRinGS-360 · fine text and texture
T squared Splat rendering of the Garden scene
Ground-truth view of the Garden scene
Ground truthT2-Splat
GardenMip-NeRF 360
T squared Splat rendering of the Truck scene
MeshSplatting rendering of the Truck scene
MeshSplattingT2-Splat
TruckTanks & Temples
T squared Splat rendering of the Bicycle scene
Ground-truth view of the Bicycle scene
Ground truthT2-Splat
BicycleMip-NeRF 360
T squared Splat rendering of the Barn scene
MeshSplatting rendering of the Barn scene
MeshSplattingT2-Splat
BarnTanks & Temples
T squared Splat rendering of the Kitchen scene
Ground-truth view of the Kitchen scene
Ground truthT2-Splat
KitchenMip-NeRF 360

Ground-truth views are reference photographs. Shelf is an additional STRinGS-360 example, separate from the evaluation averages below.

Comparisons with the State of the Art

Mesh-based novel-view synthesis results from Table 1 of the paper. Mip-NeRF 360 reports nine-scene averages; Tanks & Temples reports the Truck and Train average. All rows below use mesh-based rendering. Higher PSNR and SSIM and lower LPIPS indicate better quality.

Mesh-based novel-view synthesis
MethodMip-NeRF 360Tanks & Temples
PSNR ↑LPIPS ↓SSIM ↑PSNR ↑LPIPS ↓SSIM ↑
2DGS15.360.4740.49814.230.4850.569
GOF20.780.4650.57321.690.3260.690
RaDe-GS23.560.3610.66820.510.3440.659
MiLo24.090.3230.68821.460.3480.706
Gaussian Wrapping25.100.2890.75220.980.3670.700
Triangle Splatting21.050.4620.55817.270.4020.600
MeshSplatting24.780.3100.72820.520.2870.745
T2-Splat (Ours)25.490.2340.77921.120.2130.795

T2-Splat improves PSNR by 0.71 dB over MeshSplatting on Mip-NeRF 360. On DTU, it obtains a mean Chamfer distance of 0.68 across 15 scenes, tied for the second-best mean among the methods in Table 2.

Controlling Mesh Complexity

Changing the relative rates of edge collapse and vertex split controls the final face budget. Each T2-Splat budget configuration is trained separately with Stage I topology optimization followed by Stage II appearance refinement.

Rendering quality at different mesh budgets · Garden
MethodPSNR ↑LPIPS ↓SSIM ↑#Faces ↓
RaDe-GS22.670.210.74360.8M
GOF22.990.260.70692.1M
2DGS21.900.320.60810.6M
MiLo23.290.220.73814.9M
MeshSplatting24.700.220.7627.3M
Gaussian Wrapping23.450.270.71731.2M
T2-Splat (100%)25.680.150.8182.7M
T2-Splat (50%)25.380.170.8011.4M
T2-Splat (10%)23.930.300.6970.3M

Rendering in Standard Graphics Pipelines

The optimized mesh can be exported to standard formats. On the Garden scene, the baked vertex-colored PLY renders at 236 FPS in OpenGL at 1920 × 1080 on a MacBook Air M3, with a 62 MB mesh file. MeshSplatting reports 126 FPS and 152 MB under the same setting.

OpenGL rendering efficiency
MethodFPS ↑Storage ↓#Faces ↓
GOFOOM1.8 GB92.1M
RaDe-GSOOM1.1 GB60.8M
MiLo82314 MB14.9M
MeshSplatting126152 MB7.3M
Gaussian Wrapping40638 MB31.2M
T2-Splat (Ours)23662 MB2.7M

Vertex-colored mesh · PLY

Bake spherical harmonics and residual textures into vertex RGB. The default export preserves the original vertices, faces, and connectivity.

Textured mesh · GLB

Bake per-face RGB textures into embedded PNG atlases. The triangle count stays unchanged; face corners are expanded to support independent UVs.

Both exports use static baked appearance and do not preserve view-dependent effects. PLY can smooth fine texture details; GLB retains more intra-face detail, but its viewer mipmaps differ from training-time per-face LOD sampling. The reported FPS and storage refer to the baked PLY.

BibTeX

@inproceedings{tang2026t2splat,
  title     = {T2-Splat: Adaptive Topology Mesh Splatting with Texture Residuals},
  author    = {Tang, Zhihao and Zhang, Youjia and Zhao, Mingbo and Yang, Wei},
  booktitle = {Advances in Neural Information Processing Systems},
  year      = {2026}
}