model.py

import tensorflow as tf
from tensorflow.keras.applications import imagenet_utils
from tensorflow.keras.preprocessing.image import img_to_array
from tensorflow.keras.preprocessing.image import load_img
import numpy as np
import tensorflow.keras.backend as K
from tensorflow import keras
from keras.layers.core import Dense
from keras.layers.core import Dropout
from keras.layers.core import Flatten
from keras.models import Model
import cv2
from keras.applications import VGG16
from sklearn.preprocessing import LabelEncoder
from keras_vggface.vggface import VGGFace

def load_Pretrain_model():
    # Load model VGG 16 của ImageNet dataset, include_top=False để bỏ phần Fully connected layer ở cuối.
    model = VGGFace(include_top=False, input_shape=(224, 224, 3), pooling='avg')  # pooling: None, avg or max
    # model.summary()
    return model
def load_model():
    from tensorflow.keras.applications.resnet50 import preprocess_input
    # train_data, val_data = loadData("train",image_path,labels, val_split=0.2)
    # restnet = ResNet50(include_top=False, weights='imagenet', input_shape=(224,224,3))

    baseModel = VGGFace(include_top=False, input_shape=(224, 224, 3), pooling='avg', model='resnet50')

    # Xây thêm các layer
    # Lấy output của ConvNet trong VGG16
    fcHead = baseModel.output

    # Flatten trước khi dùng FCs
    fcHead = Flatten(name='flatten')(fcHead)

    # Thêm FC
    fcHead = Dense(256, activation='relu')(fcHead)
    fcHead = Dropout(0.5)(fcHead)

    # Output layer với softmax activation
    fcHead = Dense(429, activation='softmax')(fcHead)

    # Xây dựng model bằng việc nối ConvNet của VGG16 và fcHead
    model = model = Model(inputs=baseModel.input, outputs=fcHead)
    model.load_weights('data_model/face_recognition_restnet50.h5')
    return model
def load_model_face_detector():
    model = cv2.dnn.readNetFromCaffe("data_model/deploy.prototxt.txt",
                                     "data_model/res10_300x300_ssd_iter_140000.caffemodel")
    return model
def getFeatureImage(faceImage,model):
    crop_img = img_to_array(faceImage)
    crop_img = np.expand_dims(crop_img, axis=0)
    crop_img = imagenet_utils.preprocess_input(crop_img)
    # features = model.predict(crop_img)
    return crop_img
def load_Labels():
    label_person = []
    # open file and read the content in a list
    with open('data_model/label.txt', 'r') as filehandle:
        for line in filehandle:
            # remove linebreak which is the last character of the string
            currentPerson = line[:-1]
            # add item to the list
            label_person.append(currentPerson)
    return label_person
# Make Predictions
def cut_faces(image,faces_coord):
    (x,y,w,h) =  faces_coord
    w = w-x
    h = h-y
    w_rm = int(0.2*w/2)
    faces = (image[y:y+h,x+w_rm:x+w-w_rm])
    return faces
def resize(image,size = (224,224)):
    if image.shape < size:
        image_norm = cv2.resize(image, size, 3)
    else:
        image_norm = cv2.resize(image, size, 3)
    return image_norm
def face_detector_by_image(imagePath,model_fd,classifier_model,pretrain_model,labels):
    listPerson = []
    image = cv2.imread(imagePath)
    # print(image)
    # faces_coord = detector.detect(frame) #faces_coord is numpy array,day la toa do guong mat ma minh lay duoc
    (h, w) = image.shape[:2]
    # blobImage convert RGB (104.0, 177.0, 123.0)
    # blob = cv2.dnn.blobFromImage(image, 1.0, (224, 224), (104.0, 177.0, 123.0))
    blob = cv2.dnn.blobFromImage(cv2.resize(image, (224, 224)), 1.0, (224, 224), (104.0, 177.0, 123.0))
    model_fd.setInput(blob)
    detections = model_fd.forward()
    # print(len(faces_coord))#face_coord la mot mang luu cac toa do khuong mat tren frame
    # loop over the detections
    for i in range(0, detections.shape[2]):
        # print(j)
        box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
        (startX, startY, endX, endY) = box.astype("int")
        # extract the confidence and prediction
        confidence = detections[0, 0, i, 2]
        if (confidence > 0.2):
            # compute the (x, y)-coordinates of the bounding box for the
            # object
            face = cut_faces(image, box.astype("int"))  # tu toa do ta se cat duoc cac khuon mat

            if len(face):
                face = resize(face)
                # cv2.imwrite("maianhCrop.jpg", face)
                global name, accuracy
                img_encode = getFeatureImage(np.reshape(face, (224, 224, 3)), pretrain_model)
                number = predict(img_encode, classifier_model)
                le = LabelEncoder()
                labelsEnc = le.fit_transform(labels)
                msv = le.inverse_transform(number)
                print(msv)
                listPerson.append(msv)
                cv2.rectangle(image, (startX, startY), (endX, endY), (0, 255, 0), 2)
                cv2.putText(image, str(msv), (startX, startY - 10),
                            cv2.FONT_HERSHEY_PLAIN, 3,
                            (66, 53, 243), 2)
                # cv2.putText(image, str(accuracy), (endX, endY + 10),
                #             cv2.FONT_HERSHEY_PLAIN, 3,
                #             (66, 53, 243), 2)
    cv2.imwrite("savedImage.jpg",image)
    return listPerson
def predict(img_encode,classifier_model):
  # embed = K.eval(img_encode)
  person = classifier_model.predict(img_encode)
  y = np.argmax(person, axis=1)
  # le = LabelEncoder
  # le.inverse_transform(y)
  # le = LabelEncoder()
  # accurary = np.max(person)
  # print(le.fit_transform(person))
  return y