← all repositories
Anttwo/SuGaR

Turning unruly Gaussian splats into meshes you can rig

SuGaR wrangles the millions of disorganized Gaussians produced by 3D Gaussian Splatting into a real mesh you can rig, sculpt, and animate in standard tools.

3.4k stars C++ Computer Vision
SuGaR
Not currently ranked — collecting fresh signals.
star history

What it does SuGaR is a CVPR 2024 method that extracts a surface mesh from a 3D Gaussian Splatting scene. It adds a regularization term that nudges the Gaussians to align with the actual surface, then samples points and reconstructs a mesh using Poisson reconstruction instead of the usual Marching Cubes on neural SDFs. An optional refinement step binds Gaussians directly to that mesh, creating a hybrid representation that renders through Gaussian splatting while the underlying geometry behaves like a traditional textured mesh.

The interesting bit The regularization term is the quiet hero: it turns an intentionally unstructured cloud of millions of tiny Gaussians into something with a coherent surface, without sacrificing the speed that makes Gaussian Splatting appealing. Once aligned, the mesh can be handed off to standard DCC tools—Blender, Unity, Unreal—for rigging and animation, then rendered back through the splatting engine.

Key highlights

  • Extracts editable meshes in minutes rather than the hours typical of neural SDF pipelines, according to the authors.
  • Uses Poisson reconstruction, which the authors note is faster and more detail-preserving than Marching Cubes on neural SDFs.
  • Produces a hybrid Mesh+Gaussians representation that can be sculpted, rigged, and relit by manipulating the mesh in traditional software.
  • Ships with a dedicated real-time viewer and a Blender add-on for composition and animation.
  • A September 2024 update introduced a dn_consistency regularization method that improved extracted mesh quality.

Caveats

  • The codebase is a wrapper around vanilla 3D Gaussian Splatting that the authors admit may not be optimal for memory usage.
  • Windows is explicitly unsupported due to path-writing conventions, and several features—such as synthetic dataset support, precomputed masks, and fully cleaned composition/animation code—remain on the to-do list.

Verdict Worth exploring if you need to bridge fast Gaussian Splatting with traditional mesh-based workflows in Blender or game engines. Give it a pass if you require a polished, cross-platform tool or a lightweight, memory-optimal drop-in replacement.

Frequently asked

What is Anttwo/SuGaR?
SuGaR wrangles the millions of disorganized Gaussians produced by 3D Gaussian Splatting into a real mesh you can rig, sculpt, and animate in standard tools.
Is SuGaR open source?
Yes — Anttwo/SuGaR is an open-source project tracked on heatdrop.
What language is SuGaR written in?
Anttwo/SuGaR is primarily written in C++.
How popular is SuGaR?
Anttwo/SuGaR has 3.4k stars on GitHub.
Where can I find SuGaR?
Anttwo/SuGaR is on GitHub at https://github.com/Anttwo/SuGaR.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.