GSSA: Gaussian Surfels with Spatial Awareness for Surface Reconstruction

Abstract

3D Gaussian Splatting is effective in multi-view surface reconstruction tasks, fundamental to photogrammetry applications. Mainstream pipelines focus on optimizing for explicit Gaussians to enhance volumetric rendering quality, typically relying on rendering-based Truncated Signed Distance Function (TSDF) methods for reconstruction, while the training and meshing often suffer from missing details and geometric distortions under occlusions or restricted viewpoints. In this paper, we focus on a new problem definition: directly formulating implicit surface representation exploiting geometric features of trained Gaussian Surfels. For this problem, we introduce Gaussian Surfels with Spatial Awareness (GSSA) for reconstruction, a spatially aware reconstruction framework that optimizes for Surfel distributions from geometric primitives via a 3D refiner, and constructs SDF by integrating signed distances from voxel vertices to nearby Surfels. Compared with state-of-the-art GS-based, especially Surfel-based, reconstruction methods, GSSA achieves both competitive geometric accuracy and efficiency. On terrestrial (indoor object) and airborne (photogrammetry) reconstruction benchmarks, GSSA maintains a competitive balance between reconstruction time and accuracy. Further migration experiments demonstrate that GSSA can also be integrated as a plug-and-play module into general Surfel-based pipelines.

Method

Results

Qualitative Comparison

(a) Image Regions

(b) TSDF

(c) Poisson

(d) α-Shape

(e) GOF

(g) NeuS

(h) Ours

(i) Ours (color)

(j) Ground Truth

Qualitative comparison between GSSA and baselines on UAV and BlendedMVS datasets. Yellow boxes highlight detail regions.

Quantitative Comparison

Quantitative accuracy evaluation on UAV dataset.
Method	Precision (%) ↑	Recall (%) ↑	F1 Score (%) ↑	Chamfer ↓	Mean Normal (°) ↓	Hausdorff ↓
TSDF	91.6	92.8	92.2	0.0189	48.1	0.3274
Poisson	80.4	91.8	85.7	0.0198	39.6	0.3269
α-Shape	87.1	83.9	85.5	0.0180	48.5	0.3251
GOF	81.0	89.8	85.1	0.0199	48.8	0.3272
NeuS	81.7	77.6	79.6	0.0202	43.0	0.3149
Ours	93.4	92.8	93.1	0.0179	34.8	0.3105

Quantitative accuracy evaluation on BlendedMVS dataset.
Method	Precision (%) ↑	Recall (%) ↑	F1 Score (%) ↑	Chamfer ↓	Mean Normal (°) ↓	Hausdorff ↓
TSDF	34.4	75.6	47.3	0.0057	53.4	0.4983
Poisson	49.9	82.4	62.2	0.0051	49.4	0.7649
α-Shape	93.6	68.1	78.8	0.0011	50.8	0.2435
GOF	95.1	68.6	79.7	0.0008	43.2	0.1694
NeuS	71.9	70.1	71.0	0.0018	46.0	0.2104
Ours	88.6	77.2	82.5	0.0010	47.3	0.1683

Acknowledgements

This project benefits from 2DGS, CityGaussianV2. Thanks for their great work!

License

This project is licensed under CC BY-NC-SA 4.0. Commercial use requires prior consent. See LICENSE details.

BibTeX

If you find our work helpful, please cite:

@article{tang2026gssa,
    title = {GSSA: Gaussian Surfels with Spatial Awareness for Surface Reconstruction},
    author = {Tang, Haojun and Zou, Siyuan and Pan, Hongbo and Lu, Yixin and Zhou, Shun},
    journal = {Remote Sensing},
    year = {2026}
}