Documentation Index
Fetch the complete documentation index at: https://mintlify.com/shubham0204/OnDevice-Face-Recognition-Android/llms.txt
Use this file to discover all available pages before exploring further.
Data model classes for storing face embeddings, person records, and recognition performance metrics.
FaceImageRecord
Entity class for storing face embeddings with vector search indexing.
@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()
)
Fields
Primary key of the face image record, auto-generated by ObjectBox
Foreign key reference to PersonRecord. Indexed for efficient queries by person
Name of the person for quick access without joining PersonRecord
faceEmbedding
FloatArray
default:"floatArrayOf()"
512-dimensional face embedding vector from FaceNet model. Indexed with HNSW for efficient vector similarity search using cosine distance
Vector indexing
The faceEmbedding field uses HNSW (Hierarchical Navigable Small World) indexing:
Vector dimensionality matching FaceNet-512 model output
distanceType
VectorDistanceType
default:"COSINE"
Cosine distance metric for measuring similarity between embeddings
Usage example
// Create and store a face record
val embedding = faceNet.getFaceEmbedding(croppedFaceBitmap)
val record = FaceImageRecord(
personID = 123,
personName = "John Doe",
faceEmbedding = embedding
)
imagesVectorDB.addFaceImageRecord(record)
// Query by person ID
val personFaces = imagesBox
.query(FaceImageRecord_.personID.equal(personID))
.build()
.find()
PersonRecord
Entity class for storing person metadata and tracking enrolled faces.
@Entity
data class PersonRecord(
@Id var personID: Long = 0,
var personName: String = "",
var numImages: Long = 0,
var addTime: Long = 0
)
Fields
Primary key of the person record, auto-generated by ObjectBox
Number of face images enrolled for this person. Used for tracking enrollment completeness
Timestamp (in milliseconds) when the person was added to the database
Usage example
// Create a new person
val person = PersonRecord(
personName = "John Doe",
numImages = 0,
addTime = System.currentTimeMillis()
)
val personID = personBox.put(person)
// Update image count after enrolling faces
person.numImages = 5
personBox.put(person)
// Query all persons sorted by name
val allPersons = personBox
.query()
.order(PersonRecord_.personName)
.build()
.find()
RecognitionMetrics
Data class for tracking face recognition pipeline performance metrics.
data class RecognitionMetrics(
val timeFaceDetection: Long,
val timeVectorSearch: Long,
val timeFaceEmbedding: Long,
val timeFaceSpoofDetection: Long
)
Fields
Time in milliseconds for face detection step
Time in milliseconds for vector database nearest neighbor search
Time in milliseconds for FaceNet embedding generation
Time in milliseconds for face spoof/liveness detection
Usage example
// Metrics are automatically collected during recognition
val (metrics, results) = imageVectorUseCase.getNearestPersonName(
frameBitmap = cameraBitmap,
flatSearch = false
)
if (metrics != null) {
println("Face Detection: ${metrics.timeFaceDetection}ms")
println("Embedding Generation: ${metrics.timeFaceEmbedding}ms")
println("Vector Search: ${metrics.timeVectorSearch}ms")
println("Spoof Detection: ${metrics.timeFaceSpoofDetection}ms")
val totalTime = metrics.timeFaceDetection +
metrics.timeFaceEmbedding +
metrics.timeVectorSearch +
metrics.timeFaceSpoofDetection
println("Total: ${totalTime}ms")
}
Pipeline stages
The metrics correspond to the face recognition pipeline stages:
- Face detection: Detect and crop faces from camera frame
- Face embedding: Generate 512-dimensional vectors using FaceNet
- Vector search: Find nearest neighbor in database
- Spoof detection: Detect presentation attacks
Metrics are averaged when multiple faces are detected in a frame:
val metrics = RecognitionMetrics(
timeFaceDetection = t1.toLong(DurationUnit.MILLISECONDS),
timeFaceEmbedding = avgT2 / faceDetectionResult.size,
timeVectorSearch = avgT3 / faceDetectionResult.size,
timeFaceSpoofDetection = avgT4 / faceDetectionResult.size
)
data/DataModels.kt