I Built a Face Recognition SDK
Had an idea for an app that finds facial similarities in group selfies. Seemed simple. Wasn't.
What I Looked For First
Cloud APIs (AWS, Google): Work fine, but uploading user photos to Amazon felt wrong. Plus expensive.
Open Source: Either academic experiments with poor accuracy or server-focused with massive dependencies.
Commercial SDKs: $1k+ licenses designed for enterprise.
So I Built PerchEye SDK
Core requirements:
- Completely offline
- Cross-platform (Android, iOS, Flutter, React Native)
- Actually fast on mobile devices
- Privacy-first design
Performance achieved:
- Face detection: ~50-100ms
- Hash generation: ~150ms
- Model size: ~15MB
- Memory usage: ~75MB
- Accuracy: >95% for clear faces
Then Built Twins Finder App
Simple workflow:
- Take group selfie
- Detect faces
- Generate mathematical hashes (no actual face storage)
- Compare similarities
- Show results
Everything stays on-device. Only math representations stored, never face images.
Twins Finder - Download
PerchEye SDK - GitHub
Open Source
Released PerchEye for free because this tech shouldn't cost enterprise money.
Anyone else working on mobile computer vision? What challenges have you hit?
Top comments (0)