Practical 3D Machine Learning — A 20-Post Series

Twenty hands-on posts on loading 3D models, rendering them, embedding them, and building a working 3D similarity search engine. Code, data, and honest numbers throughout.

01. Loading a 3D Model in Python Without Crying
02. Why Your First 1,000 Renders Will Be Blank (And How to Fix It)
03. Five Ways to Render a 3D Chair in Python (and When to Pick Which)
04. Every Public 3D Dataset Worth Knowing, In One Notebook
05. Turning 3D Models Into Search Vectors With Two Lines of CLIP
06. Five Shape Descriptors, One Rotation: Who Survives?
07. Zernike Moments for 3D, From Sphere to Search Index
08. Spherical Harmonics for Shape Similarity, Without the Physics
09. When a 64-bit Hash Catches a 3D Duplicate (And When It Doesn't)
10. Train-Time Rotation Augmentation: Where the Curve Actually Flattens
11. A 3D Search Engine on $0.50 of Compute
12. Three Descriptors, One Near-Duplicate Test Suite
13. When t-SNE Lies About Your 3D Embeddings (And Sammon Tells the Truth)
14. Picking the Right Similarity Threshold When You Have No Labels
15. Stop Re-Embedding the Same Images
16. I Built a 3D Similarity Benchmark in a Weekend
17. Your Classifier Was Trained on CAD. It's About to Meet a Real Scan.
18. Prediction Sets, Not Predictions: Conformal Coverage for 3D Classifiers
19. Confidence Intervals for Classifier Scores, Borrowed From Forecasting
20. Where the Million-Image Score Job Actually Spends Its Time