HKUST-SAIL

HKUST-SAIL / Geometry-Grounded-Gaussian-Splatting

Public

Geometry-Grounded Gaussian Splatting

baowenz.github.iogeometry_grounded_gaussian_splatting
185
14
100% credibility
Found Feb 02, 2026 at 19 stars 10x -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
C++
AI Summary

An academic implementation of geometry-aware Gaussian Splatting for high-fidelity 3D reconstruction from multi-view images.

How It Works

1
🔍 Discover the tool

You find a clever way to turn everyday photos into stunning 3D models that look real and detailed.

2
📸 Gather your photos

Collect a set of pictures of your object or scene from different angles, like snapping memories on a walk.

3
🚀 Start the magic

Feed your photos into the tool and watch it study them to understand the shape and colors.

4
🛠️ Shape it into 3D

The tool carves out a solid 3D version of your scene, ready to explore from any angle.

5
✨ Behold your creation

Spin around your lifelike 3D model, seeing every detail pop just like the real thing.

🎉 Share the wonder

Export and show off your beautiful 3D reconstruction to friends or use it in your projects.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 185 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is Geometry-Grounded-Gaussian-Splatting?

Geometry-Grounded Gaussian Splatting reconstructs high-fidelity 3D scenes from multi-view images by anchoring Gaussian splatting representations to explicit geometry priors, delivering sharper novel views and extractable meshes. Users train models on datasets like DTU or Tanks & Temples via simple Python scripts, render outputs, extract watertight meshes with tetrahedral marching or TSDF fusion, and evaluate PSNR/SSIM alongside geometry metrics like Chamfer distance. Built in C++ with PyTorch rasterization, it supports decoupled appearance models from GS, GOF, or PGSR for realistic relighting.

Why is it gaining traction?

It outperforms vanilla Gaussian splatting on geometry benchmarks by integrating 3D filters, multi-view NCC losses, and densification from top papers, yielding cleaner meshes without post-processing hacks. Developers appreciate the one-command training (`python train.py`), plug-and-play eval toolkits for DTU/TnT, and SIBR-based viewer for instant previews—ideal for iterating on grounded splatting variants like VG3T visual geometry grounded Gaussian transformer.

Who should use this?

Computer vision researchers benchmarking Gaussian splatting against NeRFs on DTU/Tanks & Temples, AR/VR engineers needing geometry-accurate meshes from phone captures, or 3D reconstruction teams extending splatting with custom losses for robotics scanning.

Verdict

Solid pick for geometry-focused Gaussian splatting experiments if you're okay with its early maturity—151 stars and 1.0% credibility score mean thorough testing advised, but clear README and scripts make it accessible for quick prototypes.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.