Real-Time 3D Point Cloud Labelling from 2D Images (R&D workflow)
Code + Data → https://learngeodata.eu/tools-and-gif...
3DDS Python Book → https://www.oreilly.com/library/view/...
3D Web Viewer → https://learngeodata.eu/apps/viewers/...
Manual 3D point cloud labeling is the ultimate bottleneck in spatial AI—stealing hundreds of hours from engineers and stalling deep learning deployments. When I was leading an R&D team, watching annotators spend 6 hours fighting the camera to label a single LiDAR scan made me realize the standard industry workflow was fundamentally broken.
This session isn't about downloading another expensive, vendor-locked commercial tool. It’s about Innovation Architecture. We are building a custom, hybrid 2D-to-3D annotation GUI entirely in Python. By leveraging the geometric duality of images and point clouds, we paint masks in 2D and use a smart KD-Tree fusion algorithm to achieve geometric consensus—propagating precise labels across billions of 3D points without blowing up your RAM. This is how you stop acting as a software consumer and start engineering your own spatial intelligence systems.
⏱️ TIMESTAMPS:
[00:00] Intro & Demo
[00:38] Context: Manual AI Bottlenecks
[01:45] The 2D-to-3D Duality Approach
[03:08] Step 0: Environment & Datasets
[03:30] Installation Guide
[04:06] Verify Reconstruction Data
[04:43] Script Initialization
[05:01] Step 1: Load 3D Scene Data
[05:53] Step 2: 2D Image HD Masking
[07:28] Step 3: Projecting Masks to 3D
[08:31] Step 4: Multi-Frame Setup
[08:58] Step 5: KD-Tree Label Fusion
[09:48] Step 6: Multi-View Alignment
[10:30] The Annotation Paradigm Shift
[12:35] Hybrid Strategy
[13:56] Step 7: Analyze Scene Stats
[14:21] Step 8: Export Fused PLY Data
[14:50] 3D Engine Visualization
[15:18] Step 9: Unsupervised Segments
[16:11] Step 10: GLB & Splatting Export
[16:35] Web Visualization
[17:15] Step 11: Scale to Multi-Frame
[21:17] Human-in-the-Loop QC
[23:18] Future Perspectives
🙋 FOLLOW ME
Linkedin: / florent-poux-point-cloud
Medium: / florentpoux
#GaussianSplatting #BlenderPython #3DAI #ComputerVision #MachineLearning
WHO AM I?
If we haven’t yet before - Hey 👋 I’m Florent, a professor-turned-entrepreneur, and I’ve somehow become one of the most-followed 3D experts. Through my videos here on this channel and my writing, I share evidence-based strategies and tools to help you be better coders and 3D innovators.