The Modern Trolley Problem
summary
A modern trolley problem is reframed with a hands‑on demo: an Azure Percept device running an Azure Custom Vision object‑detection model trained on 15 images per class (demo threshold), advised to use 100+ images or semantic segmentation for production. A Lego car is built, model deployed from Custom Vision to the edge, and it reliably avoids cones but misclassifies and mows down Lego people — code and deployment scripts are available on GitHub and Microsoft Responsible AI principles guide the design choices.
AI-written summary · about
chapters (4)
related talks
33mBuilding end-to-end computer vision solutions from deep learning models
29mApplied Custom Vision
01h 33mComputer Vision in Azure
15mRobot Hand Driven by Computer Vision and Reinforcement Learning
26mSelf Driving cars with Opencv
39mNew multimodal vision AI models and their practical applications
35mComputer vision: British Clay
32mWhy Current AI Cannot Truly See the World