Spatial AI Notes & Deep Dives

Point Cloud Processing

Open-Vocabulary 3D Semantics in Python: SAM, CLIP, and DINOv2 (No Training)

Turn a careless phone video into a labeled, scaled, measurable 3D room with pretrained SAM, CLIP, and DINOv2, and not one training run. A nine-step open-vocabulary 3D semantics pipeline in Python.

26 Jun 2026 Prof. Florent Poux

Scan-to-BIM & Digital Twins

3D Spatial AI for AEC: The Offline Stack That Replaces 6 Weeks of BIM Work

From the NXT BLD 2026 stage and the reality capture panel: the local-first cognition stack for 3D spatial AI for AEC. Smart point cloud in, IFC and decisions out, in three days, not six weeks.

19 May 2026 Prof. Florent Poux

3D Computer Vision

3D Point Cloud Labeling from Photos with Python (and why AI can’t design it for you)

A complete Python pipeline for 3D point cloud labeling from phone photos using interactive painting, pinhole projection, and KD-tree fusion — no GPU required.

03 Apr 2026 Prof. Florent Poux

Spatial AI & Automation

3D Pipeline Architecture: A Founder’s Blueprint

It was a Tuesday when I got the call. A €40,000 project was completely stalled, burning cash every hour. The reason? The one person who knew the specific sequence of clicks to get data from their laser scanner software into AutoCAD was on vacation. The entire, expensive, high-tech 3D data pipeline was held hostage by […]

17 Oct 2025 Prof. Florent Poux

3D Reconstruction & Photogrammetry

Multi-View Engine for 3D Generative AI Models (Python Tutorial)

Generate perfect multi-view 3D training data without cameras. Mathematical algorithms create optimal viewpoints for Gaussian Splatting in minutes, not days. Stop Shooting Photos for 3D AI Training Data (Use This Instead) I have a fun story to share. I watched a team spend €15,000 on camera equipment and two weeks of labor to capture training […]

22 Sep 2025 Prof. Florent Poux

3D Computer Vision

3D Scene Graphs for Spatial AI with NetworkX and OpenUSD

Learn to build intelligent 3D scene graphs from point clouds using Python, NetworkX, and OpenUSD. Complete tutorial access with LLM integration for spatial AI. The hardest part of spatial AI isn’t capturing the data. It’s making that data actually intelligent. You’ve got millions of perfectly labeled 3D points, but when you try to build something […]

30 Jun 2025 Prof. Florent Poux

3D Point Cloud Labeling from Photos with Python (and why AI can’t design it for you)

By Prof. Florent Poux 15 min read Published 03 Apr 2026

A complete Python pipeline for 3D point cloud labeling from phone photos using interactive painting, pinhole projection, and KD-tree fusion…

Multi-View Engine for 3D Generative AI Models (Python Tutorial)

By Prof. Florent Poux 9 min read Published 22 Sep 2025

Generate perfect multi-view 3D training data without cameras. Mathematical algorithms create optimal viewpoints for Gaussian Splatting in minutes, not days.…