Final IronID (Android)

Offline, privacy-first gym machine identifier built with CameraX and on-device TensorFlow Lite. Captures a photo (or gallery image), runs local inference, and instantly shows instructional videos for the detected machine—including multiple variations when available.

Highlights

• On-device TFLite inference (no network needed).
• Camera capture + gallery import.
• Smart result sheet:
  • Plays instructional video for the detected machine.
  • Shows variation chips when multiple videos exist; switch instantly.
• Performance monitoring: IronID_Performance Logcat tag logs latency and confidence.
• Safety/UX:
  • Low confidence (<60%): hides video and shows toast “Confidence too low, please move closer.”
  • Permission denied: Snackbar + hint explaining camera access requirement.

Videos & Resources

• Place videos in app/src/main/res/raw/ using lowercase + underscores (e.g., bench_press.mp4).
• VideoRepository maps display labels (e.g., “Leg Press”, “Dumbbell”) to lists of R.raw.* entries; update as you add/rename files.

Building & Running

./gradlew installDebug

• Ensure a device/emulator is connected and camera permission is granted at runtime.
• If resource merge fails, verify raw filenames follow Android rules: [a-z0-9_]+.

Checking Performance Logs

adb logcat -s IronID_Performance

You’ll see entries like: Inference Latency: 50 ms - Confidence: 97.25%.

Key UX/Error-Handling Behaviors

• Low confidence (<0.60f): no video playback; toast prompts user to move closer.
• Missing camera permission: Snackbar + hint; capture disabled until granted.

Tech Stack

• Kotlin, AndroidX, CameraX, Material Components, TensorFlow Lite.