Mesh Extraction from Learned 3D Gaussian Representations

Develop a robust method to extract accurate surface meshes from trained 3D Gaussian Splatting scene representations while preserving rendering fidelity and capturing fine geometric details.

Background

Within the context of virtual human modeling, the paper notes persistent gaps between explicit 3D Gaussian representations and mesh-based outputs that are desirable for downstream applications. While several avatar methods use Gaussian splats and achieve realistic rendering, extracting clean, high-quality meshes remains challenging.

The authors explicitly identify mesh extraction from learned 3D Gaussians as future work, indicating that practical and general solutions are currently lacking.

References

ii) How to extract meshes from learned 3D Gaussians remains a future work to be investigated.

— 3D Gaussian as a New Era: A Survey  (2402.07181 - Fei et al., 2024) in Opportunities, Section 8 (Virtual Humans)

We leave improving the surface quality of the deformed mesh toward metrology-grade measurement to future work.

— CADSplat: Sparse-View 3D Gaussian Splatting Aided by CAD Models for Robust, Photorealistic Digital-Twin Reconstruction  (2609.18473 - Overdulve et al., 16 Sep 2026) in Section 6, “Limitations”

We therefore cannot presently confirm completion's benefit through geometric metrics, only its qualitative role in producing hole-free interactive scenes.

— ObjectSplat: Improving Mesh Fidelity and Interactivity for 3D Scenes via Object-Level Mesh Splatting  (2608.30423 - Kamal et al., 31 Aug 2026) in Section 4, “Limitations and Future Work”