Face detector options

Choose between MediaPipe and MLKit face detection backends

FaceNet Android supports two face detection backends: Google MLKit and MediaPipe Face Detector. Both detect faces in images and provide bounding boxes for cropping.

Available detectors

MLKit Face Detector

MLKit provides Google’s on-device face detection API with two performance modes:

• Fast mode - Used for real-time camera frame detection
• Accurate mode - Used for processing user-selected images

The MLKit detector automatically switches between modes based on the operation.

MediaPipe Face Detector

MediaPipe uses the BlazeFace short-range model for face detection. It runs in IMAGE mode and processes both camera frames and static images with the same configuration.

Switching detectors

To change the face detection backend, modify AppModule.kt:15:

AppModule.kt

@Module
@ComponentScan("com.ml.shubham0204.facenet_android")
class AppModule {

    // Set to true for MLKit, false for MediaPipe
    private var isMLKit = true

    @Single
    fun provideFaceDetector(context: Context): BaseFaceDetector = if (isMLKit) {
        MLKitFaceDetector(context)
    } else {
        MediapipeFaceDetector(context)
    }
}

MLKit is enabled by default and provides better performance on most devices through its dual-mode approach.

MLKit configuration

The MLKit detector creates two detector instances in MLKitFaceDetector.kt:20-28:

MLKitFaceDetector.kt

// For real-time camera detection
private val realTimeOpts = FaceDetectorOptions.Builder()
    .setPerformanceMode(FaceDetectorOptions.PERFORMANCE_MODE_FAST)
    .build()
private val realTimeFaceDetector = FaceDetection.getClient(realTimeOpts)

// For static image processing
private val highAccuracyOpts = FaceDetectorOptions.Builder()
    .setPerformanceMode(FaceDetectorOptions.PERFORMANCE_MODE_ACCURATE)
    .build()
private val highAccuracyFaceDetector = FaceDetection.getClient(highAccuracyOpts)

Customizing MLKit options

You can configure additional MLKit options:

val options = FaceDetectorOptions.Builder()
    .setPerformanceMode(FaceDetectorOptions.PERFORMANCE_MODE_ACCURATE)
    .setLandmarkMode(FaceDetectorOptions.LANDMARK_MODE_ALL)
    .setClassificationMode(FaceDetectorOptions.CLASSIFICATION_MODE_ALL)
    .setMinFaceSize(0.15f)  // Minimum face size relative to image
    .enableTracking()  // Track faces across frames
    .build()

Enabling additional features like landmarks or classification increases processing time and is not required for face recognition.

MediaPipe configuration

The MediaPipe detector uses the BlazeFace short-range model in MediapipeFaceDetector.kt:23-31:

MediapipeFaceDetector.kt

private val modelName = "blaze_face_short_range.tflite"
private val baseOptions = BaseOptions.builder()
    .setModelAssetPath(modelName)
    .build()
private val faceDetectorOptions = FaceDetector.FaceDetectorOptions
    .builder()
    .setBaseOptions(baseOptions)
    .setRunningMode(RunningMode.IMAGE)
    .build()
private val faceDetector = FaceDetector.createFromOptions(context, faceDetectorOptions)

Customizing MediaPipe options

You can adjust MediaPipe detection parameters:

val options = FaceDetector.FaceDetectorOptions
    .builder()
    .setBaseOptions(baseOptions)
    .setRunningMode(RunningMode.IMAGE)
    .setMinDetectionConfidence(0.5f)  // Detection threshold
    .setMinSuppressionThreshold(0.3f)  // NMS threshold
    .build()

Performance comparison

Detector	Camera Frame Detection	Static Image Detection	Model Size
MLKit Fast	~15-25ms	N/A	Bundled with Google Play Services
MLKit Accurate	N/A	~30-50ms	Bundled with Google Play Services
MediaPipe	~20-35ms	~20-35ms	2.8 MB

MLKit relies on Google Play Services and may download models on first use. MediaPipe bundles the model in your APK, increasing app size but ensuring offline availability.

When to use each detector

Use MLKit when:

• You need the fastest real-time detection
• Your app targets devices with Google Play Services
• You want to minimize APK size

Use MediaPipe when:

• You need guaranteed offline functionality
• You’re targeting devices without Google Play Services
• You want consistent performance across modes