estebanuri

def pre_process(face, required_size=(160, 160)):

    ret = cv2.resize(face, required_size)
    #ret = cv2.cvtColor(ret, cv2.COLOR_BGR2RGB)
    ret = ret.astype('float32')
    # standardize pixel values across channels (global)
    mean, std = ret.mean(), ret.std()
    ret = (ret - mean) / std

    return ret

estebanuri

import cv2
import matplotlib.pyplot as plt
from mtcnn.mtcnn import MTCNN

def read_image(file):
  img = cv2.imread(file)
  img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
  return img

def crop_bb(image, detection, margin):

estebanuri

  
// changes made to the model implementation
public class TFLiteObjectDetectionAPIModel
        implements SimilarityClassifier {
  
  ...
  
  // the output size is 192
  private static final int OUTPUT_SIZE = 192;
  

estebanuri

// Changes made to the model interface
// This interface was renamed from Classifier to SimilarityClassifier
public interface SimilarityClassifier {

  // added method for registering samples into the dataset
  void register(String name, Recognition recognition);
  ...
  
  public class Recognition {

estebanuri

import tensorflow.lite as lite

input_file = "models/sandberg/facenet_keras.h5"
output_file = "models/sandberg/facenet.tflite"

# Converts the Keras model to TensorFlow Lite
converter = lite.TocoConverter.from_keras_model_file(input_file)
converter.post_training_quantize = True
tflite_model = converter.convert()
open(output_file, "wb").write(tflite_model)

estebanuri

import numpy as np
from tensorflow.keras import models

# Loads Hiroki's FaceNet model
model_file = 'models/hiroki/facenet_keras.h5'
hiroki_model = models.load_model(model_file)

# Creates a random image
img = np.random.randint(0, 255, (160, 160, 3), dtype='uint8')

estebanuri

import tensorflow.lite as lite

input_file = "models/hiroki/facenet_keras.h5"
output_file = "models/hiroki/facenet.tflite"

# Converts the Keras model to TensorFlow Lite
converter = lite.TocoConverter.from_keras_model_file(input_file)
converter.post_training_quantize = True
tflite_model = converter.convert()
open(output_file, "wb").write(tflite_model)

estebanuri

 @Override
  public List<Recognition> recognizeImage(final Bitmap bitmap) {
   
    ...

    Object[] inputArray = {imgData};
  
// Here outputMap is changed to fit the Face Mask detector
    Map<Integer, Object> outputMap = new HashMap<>();
    output = new float[1][2];

estebanuri

public class DetectorActivity 
  extends CameraActivity 
  implements OnImageAvailableListener {

  ...
    
  // Face Mask
  private static final int TF_OD_API_INPUT_SIZE = 224;
  private static final boolean TF_OD_API_IS_QUANTIZED = false;
  private static final String TF_OD_API_MODEL_FILE = "mask_detector.tflite";

estebanuri

private void onFacesDetected(long currTimestamp, List<Face> faces) {

  
    final List<Classifier.Recognition> mappedRecognitions =
            new LinkedList<Classifier.Recognition>();

  ...

    
    final Canvas cvFace = new Canvas(faceBmp);