Real-Time Object Detection with CoreML
summary
A workflow for building custom on-device detectors: use Cloud Annotations to upload and collaboratively label images, run the Kackley CLI to push data into Watson Machine Learning and train on IBM GPUs, then export Core ML, TFJS and TFLite models. Recommendations: start with >=50 images, train ~20× steps per image, expect ~30 minutes on cloud GPUs, augment for rotation/scale, and watch for dataset bias (day/night). Demos include a beer and emoji detector that mostly work but fail under poor lighting.
AI-written summary · about
chapters (4)
- 0:00Introduction to object detection
- 10:10Training phases and CoreML
- 16:01Live demo with Cloud Annotations
- 29:46Q&A on model training
related talks
30mLearning CoreML with fun
55mCore ML: Vision
30mUnderstanding What Artificial Intelligence Actually Sees
49mKeynote: From Zero to ML on Google Cloud Platform
40mDeep Learning on iOS
42mCreating Smarter Applications
21mIntroduction to object detection with Vertex AI and ML Kit
38mDeep Learning for detection on a phone