zju3dv/PGSR

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1092 stars · 95 forks observed · 2026-08-28

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

PGSR is a Python implementation of a planar-based Gaussian Splatting method for efficient, high-fidelity 3D surface reconstruction from multi-view RGB images, published at TVCG 2024. It renders unbiased depth and normal maps from Gaussian primitives with geometric regularization to produce accurate meshes without requiring depth or normal priors.

Use cases

• reconstruct 3D meshes from multi-view RGB photos
• surface reconstruction from DTU or Tanks and Temples datasets
• get high-fidelity geometry from Gaussian splatting scenes
• compare against NeRF and 3DGS-based reconstruction baselines
• reconstruct indoor and outdoor scenes with illumination variation
• fast mesh extraction from trained 3D Gaussian scenes

When to choose

• you need accurate meshes from multi-view images without depth or normal priors
• you want faster training than NeRF-based surface reconstruction methods
• you need both high-quality rendering and geometric accuracy from one representation
• you are benchmarking surface reconstruction on DTU, Tanks and Temples, or Mip-NeRF 360

When to avoid

• you only need novel view synthesis without mesh geometry
• you lack a CUDA-capable GPU
• you need a permissively licensed production-ready library
• you want a turnkey application with a GUI rather than a research codebase

Facets

library · maturity active

graphics image-processing machine-learning simulation computer-vision graphics machine-learning python cross-platform 3dgs gaussian-splatting surface-reconstruction mesh-reconstruction multi-view-stereo nerf computer-graphics research-code 3d-reconstruction research linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="zju3dv/PGSR")