From concept to production: design and optimize computer vision applications on Arm
summary
Roadmap from data to deployed vision app that actually runs on Arm hardware: use on-prem data collection and pixel-level annotation for domain-specific cases like helmet detection across thousands of cameras, augment with synthetic GAN data, train with transfer learning, then quantize and prune and retrain the model before measuring accuracy and latency on target boards using the Arm NN SDK and Ethos NPU. Package with Docker, version models per device, and deploy an active-learning loop that samples frames for continuous retraining.
AI-written summary · about
chapters (5)
related talks
28mML on the Edge for Industry 4.0 with Arm Ethos N-78 Neural Processing Unit
42mThe Stakes of Green AI for Embedded Systems
20mArchitecting AI: Efficient Models, Everywhere
26mWhere can I park? Deep Learning is here to help you!
29mApplied Custom Vision
46mComputer Vision to Secure Your Surroundings with AI/ML Solution Built Using Open Source Tools at the Edge
18mHow to turn your ideas in AI into products
32mComputer Vision IRL: From Idea to On-Premise Deployment