3D Data Science with Python

Chapter 14. Supervised 3D Machine Learning Fundamentals

We aim to design machines that can perceive space precisely as humans do. This means understanding depth, predicting spatial relationships, recognizing objects, and making intelligent decisions in digital environments. This is a complex goal but more manageable when broken into chunks. In previous chapters, we solved the first two chunks: understanding depth and finding spatial relationships. But what about recognizing objects with a clear concept attached to them?

This could allow us to build systems that analyze medical images to detect anomalies like tumors, helping physicians formulate diagnostic and treatment plans. Beyond healthcare, self-driving cars rely heavily on 3D perception to navigate complex environments safely. We can create autonomous driving assistance by identifying obstacles, pedestrians, and traffic signs using LiDAR sensors and cameras. Again, every 3D data science application detailed in Chapter 1 relies on semantic extraction.

However, transforming complex 3D data into meaningful insights is the fundamental challenge. 3D machine learning, with supervised learning models, brings a critical paradigm shift in this context.

Point clouds, depth maps, meshes, and volumetric representations all contain intricate information that classical algorithms struggle to interpret. We previously saw that unsupervised learning can help us “decode” the overall meaning of our 3D environments. This is a first step toward ...