World Models & Spatial AI
3D Forge: Build Camp
Install one app, then build four working 3D AI tools end to end. Text-prompted segmentation, frugal geometric segmentation, video to floor plan, and point cloud to digital twin. You read and modify the real code, not slides.
Start here
Your onboarding
A short tour of the Forge and the four systems you will build, so you know exactly where to start.
Note from Florent
I’m tired of tutorials that stop at “run the training script.” Build Camp is the opposite. You install the app once, then you open four real systems I use, you read the exact file and function behind each step, and you change them on your own data. By the end you’ll have shipped a segmentation, a floor plan, and a queryable twin, and you’ll know why each one works.
What you will build
Module 01
Orientation, get the Forge running
One app, one install. You launch the studio, learn the GPU vs CPU matrix, and tour where the code lives so nothing blocks you.
Module 02
Text-Prompted 3D Segmentation
Type a sentence, get masks in 3D. Render to 2D, run SAM and CLIP, back-project and vote. No annotation, no training.
Module 03
Unsupervised Geometric Segmentation (Frugal)
Pure geometry splits the cloud on a laptop. Region growing and RANSAC planes plus DBSCAN, the PhD lineage productized.
Module 04
Video to 3D Splat to Floor Plan
Walk a space with a phone. Get a dense cloud, an optional Gaussian splat, and a measured floor plan you export to SVG, PDF, or DXF.
Module 05
Point Cloud or Splat to Digital Twin
Turn any cloud or splat into a game-ready mesh, a voxel model, or a queryable scene graph with per-class instances.
Module 06
Community, Challenges and What’s Next
Ship from your own data, add your own workflow node, and see where the 3D Spatial Architect program takes you next.
Your starting resources
Additional guides and code packs open up as you progress through each module.
