Navigating the Machine Learning Landscape for Apple Devices
summary
Shows how to detect the game box with Vision, crop and classify Agricola pieces with Create ML, export a 66 KB Core ML model and run inference with Core ML, and do style transfer with Turi Create. Data collection uses phone videos + ffmpeg to extract frames (~100 images), uses an 80/20 train/test split, Create ML reaches 100% train/validation accuracy and 96% test accuracy, and style transfer trained for 300 iterations on CPU (about 3 hours) and exports as .mlmodel. Also mentions weight quantization to 16/8 bits.
AI-written summary · about
chapters (10)
- 0:00Introduction to ML on Apple
- 5:18Project overview
- 6:36Using Vision for detection
- 9:16Training models with Create ML
- 14:02Data collection and preparation
- 18:05Evaluating model accuracy
- 23:28Exporting and optimizing models
- 26:25Integrating Core ML into apps
- 30:37Style transfer with Turi Create
- 39:11Conclusion
