Mobile · on-device ML
On-device computer vision
Mobile developer
Not many full-stack developers ship computer vision. On mobile, I used Google ML Kit for real-time face detection to support identity verification, and object detection plus image labeling to validate vehicle photos before upload. Everything runs on-device, with a graceful fallback for when ML isn't available.
- Flutter
- Google ML Kit
- Face detection
- Object detection
- Image labeling
- Runtime
- On-device
- Verification
- Real-time face detection
- Validation
- Object detection + labeling
- Platform
- Mobile
On-device / ZEVA Vision
The image never leaves the phone.
Camera input, face detection for identity verification, and object detection plus labeling for photo validation all run through Google ML Kit on the device itself.
A routine check needs no network round trip, and there is a graceful fallback for when ML is not available.
Nothing leaves the device for a check that can be answered on it.
The problem
Identity and photo checks had to happen in real time, on the device, without sending raw images to a server for every frame.
What I owned
I integrated Google ML Kit on-device: real-time face detection for identity verification, and object detection plus image labeling to check that submitted vehicle photos matched what was expected.
The hard parts
Running detection in real time on-device without degrading the experience.
I handled camera input on the device and ran detection inside the capture loop, so the preview stays responsive while it works.
Turning noisy detection results into a clear pass or fail for users.
I resolved detection and labeling into one verdict a person can act on, rather than a stream of confidence scores.
Keeping image handling on-device for privacy.
Nothing is uploaded for a routine check, and the path degrades gracefully when ML is unavailable.
The result
I shipped on-device vision features that verify identity and validate photos in real time.