What it does

This repo is a curated set of small, single-purpose iOS apps demonstrating how to run machine learning inference on iPhone using Core ML, ML Kit (TensorFlow Lite), and occasionally PyTorch Mobile. Each sub-project tackles one task—image classification, object detection, pose estimation, depth prediction, semantic segmentation—so you can copy the pattern rather than reverse-engineer Apple’s documentation.

The interesting bit

The author didn’t just dump links. There’s a performance table measured on an iPhone X showing inference time, total execution time, and FPS for each demo. A few projects even include a built-in “Measure” overlay so you can see latency numbers live on screen. It’s the kind of obsessive benchmarking that saves you from discovering your “real-time” model runs at 1 FPS in production.

Key highlights

• 10+ baseline projects covering vision tasks from MobileNet classification to SSD object detection to hourglass pose estimation
• Side-by-side Core ML and ML Kit implementations for several tasks, making framework trade-offs concrete
• Includes a Create ML walkthrough and a custom keypoint annotation tool for building your own datasets
• Performance comparison table with actual millisecond timings (e.g., DepthPrediction-CoreML: 624ms inference, 1 FPS)
• Most demos include GIFs showing the running app

Caveats

• Several items in the TODO list remain unfinished, including ML Kit object detection and anything involving NLP or audio
• Unit tests and batch tests are largely absent; only ObjectDetection-CoreML and PoseEstimation-CoreML have unit tests
• Some performance cells are blank (FaceDetection-MLKit has no numbers)

Verdict

Grab this if you’re an iOS developer who needs to ship a Core ML model this week and wants working starter code. Skip it if you’re looking for production-grade architecture, comprehensive test coverage, or guidance on model training.