Skip to content
imaged
Docs / Setup / Kotlin

Kotlin setup.

Same AAR as Java, idiomatic Kotlin call sites. CameraX-ready in fifty lines.

Prerequisites
Android Gradle Plugin 8.2 or later
minSdk 24 (Android 7.0)
Kotlin 2.0 or later
License token JWT issued from /dashboard
  1. 01

    Download the AAR

    Pull imaged.aar from the downloads page. Place it in app/libs/. arm64-v8a is the production target; x86_64 is included for emulators.

    bash 
    mkdir -p app/libs
cp ~/Downloads/imaged.aar app/libs/
  2. 02

    Wire up Gradle

    Mirror the example app's build.gradle.kts: pull AARs from libs, keep ONNX files uncompressed, and use legacy JNI packaging so libsdk.so is extracted on install.

    kotlin 
    // app/build.gradle.kts
android {
    namespace = "com.acme.detector"
    compileSdk = 36
    defaultConfig { minSdk = 24 }

    androidResources {
        @Suppress("UnstableApiUsage")
        noCompress += "onnx"
    }
    packaging {
        jniLibs { useLegacyPackaging = true }
    }
}

dependencies {
    implementation(fileTree("libs") { include("*.aar") })
}
  3. 03

    Hello world

    Initialize the loader once per process, construct ImagedSDK, set the license, verify it. ImagedLoader.load(context) also stages the Hexagon skel files so QNN HTP can engage on Snapdragon.

    kotlin 
    import dev.imaged.sdk.ConfigurationOptions
import dev.imaged.sdk.ImagedLoader
import dev.imaged.sdk.ImagedSDK

class Detector(context: android.content.Context) {
    private val ai: ImagedSDK

    init {
        ImagedLoader.load(context)
        ai = ImagedSDK()
        val cfg = ai.options
        cfg.setString("license.token", "your-license-token")
        cfg.setString("license.audience", "cameras.acme.com")

        val status = cfg.checkLicense()
        check(status == ConfigurationOptions.STATUS_SUCCESS && ai.isLicensed) {
            "license check failed, status $status"
        }
        android.util.Log.i("imaged", "version ${ai.version}")
    }
}
  4. 04

    Run a family

    Image implements AutoCloseable, so use { } closes the native handle for you. detect returns a ResponseObject; if response.error is non-zero, log message and skip the frame.

    kotlin 
    import dev.imaged.sdk.ColorCode
import dev.imaged.sdk.Image

ai.options.setString("yolo.model", modelPath)
ai.options.setString("yolo.inference.provider", "qnn")

Image().use { img ->
    img.setData(rgbaBytes, width, height, ColorCode.RGBA)
    val response = ai.detect("yolo", img)
    response.objects.orEmpty().forEach { obj ->
        android.util.Log.i("imaged", "${obj.label} ${obj.score}")
    }
}
  5. 05

    Where to go next

    CameraDetector.kt in the Android example is the reference. It runs CameraX, feeds RGBA frames into the SDK on a single-thread executor, and posts results to a Compose overlay.

Working example
ObjectDetector-android
Where to go next
Setup for another language
C C++ Java Objective-C Python Swift