microg/GmsCore · critical · RuntimeException

faceDetectorYN initialization failed

Error message

faceDetectorYN initialization failed

What it means

FaceDetectorHelper's constructor loads an ONNX face-detection model from assets and calls OpenCV FaceDetectorYN.create; any exception during model load/creation is rethrown as this RuntimeException. It means the bundled DNN model could not be initialized, so face detection is unusable.

Solutions

1. Verify the onnx model file exists in src/main/assets (or wherever `assets.open` reads it) and is not truncated.
2. Ensure your OpenCV dependency is >= 4.5.1 and includes the dnn module.
3. Log the original exception (`throw RuntimeException("faceDetectorYN initialization failed", e)`) to see the root cause before blaming config.
4. Reduce/validate inputSize and try the default 300x300; test FaceDetectorYN.create in isolation.

Example fix

// before
} catch (e: Exception) {
    throw RuntimeException("faceDetectorYN initialization failed")
}
// after
} catch (e: Exception) {
    throw RuntimeException("faceDetectorYN initialization failed", e)
}

val model = assets.list("")?.contains("onnx") ?: false
if (!model || !OpenCVLoader.initDebug()) failFast("OpenCV/model unavailable")

try {
    val detector = FaceDetector(context)
} catch (e: RuntimeException) {
    Log.e(TAG, "Face detection unavailable: ${e.message}", e)
    disableFeature()
}

When it happens

Trigger: Constructing FaceDetector (which instantiates FaceDetectorHelper) when the model asset is missing/corrupt, asset read fails, inputSize is invalid, or OpenCV's DNN module can't parse the onnx model.

Common situations: OpenCV dependency without dnn/onnx support; corrupt or truncated asset file; too-large inputSize for available memory; mismatched OpenCV version lacking FaceDetectorYN.

Related errors

• Access denied, missing google package permission for
• DroidGuard should not be available locally
• ERROR_MISSING_INSTANCEID_SERVICE
• Failed to obtain login token.
• Failed to retrieve client information.

AI-assisted analysis of microg/GmsCore@157c9d86ac (2026-09-06). Data as JSON: /api/errors/997e25fd9723cf41. Report an issue: GitHub.

Appendix: source

Thrown at play-services-vision/core/src/main/kotlin/org/microg/gms/vision/face/FaceDetectorHelper.kt:50

const val TAG = "FaceDetection"

class FaceDetectorHelper(context: Context) {

    private var faceDetectorYN: FaceDetectorYN? = null
    private var inputSize = Size(320.0, 320.0)

    init {
        try {
            val buffer: ByteArray
            context.assets.open("face_detection_yunet_2023mar.onnx").use {
                val size = it.available()
                buffer = ByteArray(size)
                it.read(buffer)
            }
            faceDetectorYN = FaceDetectorYN.create("onnx", MatOfByte(*buffer), MatOfByte(), inputSize, 0.7f, 0.3f, 5000)
        } catch (e: Exception) {
            throw RuntimeException("faceDetectorYN initialization failed")
        }
    }

    fun detectFaces(bitmap: Bitmap, rotation: Int): List<FaceParcel> {
        Log.d(TAG, "detectFaces: source is bitmap")
        val rootMat = bitmapToMat(bitmap) ?: return emptyList()
        return processMat(rootMat, rotation)
    }

    fun detectFaces(nv21ByteArray: ByteArray, width: Int, height: Int, rotation: Int): List<FaceParcel> {
        Log.d(TAG, "detectFaces: source is nv21Buffer")
        val rootMat = nv21ToMat(nv21ByteArray, width, height) ?: return emptyList()
        return processMat(rootMat, rotation)
    }

    fun detectFaces(image: Image, rotation: Int): List<FaceParcel> {
        Log.d(TAG, "detectFaces: source is image")
        val rootMat = imageToMat(image) ?: return emptyList()

View on GitHub (pinned to 157c9d86ac)