To recognize faces, you first need to build a database of known faces. FaceNet Android allows you to select images from your gallery and label them with person names.
How it works
When you add faces to the database, the app performs the following steps:
- Face detection: MLKit or Mediapipe detects and crops faces from your selected images
- Embedding generation: The FaceNet model converts each cropped face into a 512-dimensional (or 128-dimensional) vector embedding
- Storage: Face embeddings are stored in an ObjectBox vector database for fast nearest-neighbor search
Adding a new person
Open the face list
From the main camera screen, tap the face icon in the top-right corner to open the Face List screen.
Start adding a face
Tap the floating action button (+ icon) to navigate to the Add Face screen.
Enter the person's name
In the text field, enter the name of the person whose face you want to add to the database.TextField(
modifier = Modifier.fillMaxWidth(),
value = personName,
onValueChange = { personName = it },
label = { Text(text = "Enter the person's name") },
singleLine = true,
)
Select photos
Tap the “Choose photos” button to open the photo picker. You can select multiple images of the same person.For better recognition accuracy, select multiple photos of the person from different angles and lighting conditions.
Add to database
Review the selected images in the grid, then tap “Add to database”. The app will process each image:
- Detect faces in the image
- Generate embeddings using the FaceNet model
- Store the embeddings in the database
A progress dialog shows how many images have been processed. Under the hood
The AddFaceScreenViewModel handles the image processing workflow:
AddFaceScreenViewModel.kt
fun addImages() {
isProcessingImages.value = true
CoroutineScope(Dispatchers.Default).launch {
val id = personUseCase.addPerson(
personNameState.value,
selectedImageURIs.value.size.toLong(),
)
selectedImageURIs.value.forEach {
imageVectorUseCase
.addImage(id, personNameState.value, it)
.onFailure {
val errorMessage = (it as AppException).errorCode.message
setProgressDialogText(errorMessage)
}.onSuccess {
numImagesProcessed.value += 1
setProgressDialogText("Processed ${numImagesProcessed.value} image(s)")
}
}
isProcessingImages.value = false
}
}
Database structure
Face data is stored in two entities:
@Entity
data class FaceImageRecord(
@Id var recordID: Long = 0,
@Index var personID: Long = 0,
var personName: String = "",
@HnswIndex(
dimensions = 512,
distanceType = VectorDistanceType.COSINE,
) var faceEmbedding: FloatArray = floatArrayOf(),
)
@Entity
data class PersonRecord(
@Id var personID: Long = 0,
var personName: String = "",
var numImages: Long = 0,
var addTime: Long = 0,
)
- PersonRecord: Stores person metadata (name, number of images, timestamp)
- FaceImageRecord: Stores individual face embeddings with HNSW indexing for fast vector search
Managing faces
From the Face List screen, you can:
- View all people in your database with timestamps
- Remove people by tapping the X icon next to their name
- Add more people using the floating action button
Removing a person deletes all their face embeddings from the database. This action cannot be undone.
Next steps
Once you’ve added faces to the database, you can: