cvlab-kaist

cvlab-kaist / GLD

Public

Official implementation of "Repurposing Geometric Foundation Models for Multi-view Diffusion"

85
2
100% credibility
Found Mar 25, 2026 at 85 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Geometric Latent Diffusion is an open-source research project that generates novel views and 3D reconstructions from multi-view images using geometric AI models.

How It Works

1
🔍 Discover GLD

You find this cool project that turns a few photos into new viewpoints and 3D models.

2
💻 Get ready

Follow simple steps to prepare your computer so everything runs smoothly.

3
📥 Grab smart models

Download ready-to-use brainpower that understands shapes and scenes.

4
▶️ Run the demo

Click to launch and watch it create new angles from sample photos.

5
✨ Magic happens

New views appear and a colorful 3D scene builds itself before your eyes.

6
👀 Explore your 3D world

Open the scene in a free viewer and rotate around your creation.

🎉 You've got 3D superpowers

Now you can recreate scenes from photos anytime with realistic new views.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 85 to 85 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 GLD?

GLD repurposes geometric foundation models like Depth Anything 3 and VGGT for multi-view diffusion, generating consistent novel views from sparse camera inputs without text conditioning. It delivers zero-shot depth maps and 3D geometry straight from latents, solving slow convergence in traditional NVS pipelines. Python repo with pretrained models on Hugging Face, demo script for instant RGB+3D exports via GLB.

Why is it gaining traction?

4.4x faster training than VAE methods, SOTA scores on RealEstate10K/DL3DV benchmarks. One-command demos produce COLMAP-compatible outputs; cascade mode refines coarse-to-fine latents for better geometry. Official GitHub repository with CLI scripts for train/eval, pulling eyes from CV circles tracking goldpreis-like efficiency gains.

Who should use this?

Computer vision researchers benchmarking diffusion for NVS. 3D reconstruction engineers needing geometry priors without custom VAEs. KAIST/Intel-affiliated teams prototyping multi-view synthesis on CUT3R datasets.

Verdict

Solid paper implementation worth forking for experiments (85 stars), but 1.0% credibility signals early-stage—docs solid, tests sparse. Grab official GitHub releases if you've got 48GB VRAM; skip for production until gld_anni stabilizes.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.