city-super/Scaffold-GS

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1278 stars · 124 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Scaffold-GS is a research implementation of a structured 3D Gaussian splatting method that uses anchor points on a sparse voxel grid to distribute local 3D Gaussians, predicting their attributes on-the-fly based on viewing direction and distance. It achieves faster convergence, fewer primitives, and better visual quality than vanilla 3D Gaussian splatting, especially on scenes with challenging views.

Use cases

• reconstruct 3D scenes from photos with gaussian splatting
• render novel views of a captured scene
• improve rendering quality on reflective or transparent surfaces
• train a view-adaptive 3D gaussian model from SfM points
• compare structured gaussian splatting against vanilla 3DGS
• visualize trained scenes with the SIBR viewer

When to choose

• you need high-quality novel view synthesis with fewer gaussians and faster convergence
• your scenes contain transparency, specularity, reflection, or fine-scale detail
• you are doing research on gaussian splatting and want an anchor-based baseline
• you have a CUDA-capable Linux GPU environment for training

When to avoid

• you need a production-ready, well-documented rendering library rather than research code
• you cannot use CUDA or a Linux-based training setup
• you need real-time rendering on low-end hardware without a GPU
• you want a simple off-the-shelf viewer without compiling SIBR_viewers

Facets

library · maturity active

graphics image-processing machine-learning deep-learning computer-vision graphics machine-learning python cpp gaussian-splatting 3d-reconstruction novel-view-synthesis cvpr2024 neural-rendering research-code research linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="city-super/Scaffold-GS")