☘️ Klover — Four-Leaf Clover Finder

A Kotlin Multiplatform (Android + iOS) app that finds four-leaf clovers with on-device AI and highlights them in a live camera feed.

FOUR-LEAF-style dark, camera-first UI. The home screen is a real-time camera scanner: a live preview with detection boxes, % confidence chips, and an fps counter. Each frame runs on-device inference — ONNX Runtime on Android (CameraX), CoreML/Vision on iOS (AVCaptureSession); if no model is bundled it falls back to MockCloverDetector. The UI is 100% shared via Compose Multiplatform.

Status

Feature	Android	iOS
Shared UI (Material 3 Expressive · dark FOUR-LEAF theme)	✓	✓
Real-time camera scanner (home)	✓ CameraX + per-frame inference	✓ AVCaptureSession + per-frame inference
On-device inference	✓ ONNX Runtime (~10 fps)	✓ CoreML / Vision
YOLO post-processing (decode + NMS)	✓ shared	✓ shared

The bundled model (shared/src/commonMain/composeResources/files/clover.onnx) is a YOLOv8 model trained on public four-leaf-clover datasets — see training/ to (re)train your own for free.

Tech stack

• Kotlin Multiplatform + Compose Multiplatform (shared UI), Material 3 Expressive
• Koin (DI), AndroidX Lifecycle ViewModel (state), Coroutines
• ONNX Runtime (Android) / CoreML + Vision (iOS) for on-device inference
• AGP 9 / Kotlin 2.4 / Compose MP 1.11 / Gradle 9.6

Running

Prerequisites

• JDK 17, Android SDK (compileSdk 37), Xcode (for iOS)
• sdk.dir in local.properties (already generated locally)

Android

./gradlew :androidApp:assembleDebug     # build APK
./gradlew :androidApp:installDebug      # install on a connected device/emulator

Or run the androidApp configuration from Android Studio. A physical device is recommended so the camera works.

iOS

open iosApp/iosApp.xcodeproj            # run from Xcode

For device builds, copy iosApp/Configuration/Local.xcconfig.template to Local.xcconfig and set your Apple TEAM_ID there (it's gitignored). Bundle id: io.github.klover.

Verify the shared framework compiles:

./gradlew :shared:linkDebugFrameworkIosSimulatorArm64

Train your own model (free)

See training/README.md. In short: open the Colab notebook, download a public Roboflow clover dataset, train YOLOv8, export clover.onnx, and drop it into shared/src/commonMain/composeResources/files/. No server, no cost.

If a multi-class model is used (e.g. 3/4/5-leaf), set the 4-leaf class index in YoloConfig.TARGET_CLASS_INDEX.