What it does Reads China’s second-generation ID cards using on-device machine learning. The app takes a photo, runs it through image processing, and extracts text fields. It was built for iOS 11 and Xcode 9, back when CoreML was fresh enough to mention in a changelog.

The interesting bit The author swapped out a hand-rolled FFNN (feed-forward neural network) for Apple’s then-new Vision + CoreML stack in September 2017. That’s a neat snapshot of the iOS ML transition era — from custom nets to Apple’s blessed pipeline. The README also candidly admits the model struggles with low-quality images “due to computer performance issues,” which is charmingly honest.

Key highlights

• On-device OCR, no cloud round-trip required
• Replaced custom neural net with Apple’s Vision/CoreML in 2017
• Includes a visual pipeline showing image processing stages
• Apache 2.0 licensed
• Author migrated to a newer multi-card OCR project (Evil)

Caveats

• Explicitly deprecated; author points to new repo
• Model accuracy drops on poor-quality images
• TODO items left unfinished: training data generation, *.h5 to *.mlmodel conversion
• Locked to iOS 11 / Xcode 9 era; unlikely to build cleanly on modern toolchains

Verdict Worth a quick look if you’re studying early CoreML adoption patterns or Chinese ID card OCR archaeology. Skip it if you need something that builds today — follow the deprecation link to Evil instead.