dataset # import the necessary packages

packages from tensorflow.keras.preprocessing.image import ImageDataGenerator from tensorflow.keras.applications import MobileNetV2 from tensorflow.keras.layers import AveragePooling2D from tensorflow.keras.layers import Dropout from tensorflow.keras.layers import Flatten from tensorflow.keras.layers import Dense from tensorflow.keras.layers import Input from tensorflow.keras.models import Model from tensorflow.keras.optimizers import Adam from tensorflow.keras.applications.mobilenet_v2 import preprocess_input from tensorflow.keras.preprocessing.image import img_to_array from tensorflow.keras.preprocessing.image import load_img from tensorflow.keras.utils import to_categorical from sklearn.preprocessing import LabelBinarizer from sklearn.model_selection import train_test_split from sklearn.metrics import classification_report from imutils import paths import matplotlib.pyplot as plt import numpy as np import argparse import os

os # construct the argument parser and parse the arguments

arguments ap = argparse.ArgumentParser() ap.add_argument("-d", "--dataset", required=True, help="path to input dataset") ap.add_argument("-p", "--plot", type=str, default="plot.png", help="path to output loss/accuracy plot") ap.add_argument("-m", "--model", type=str, default="mask_detector.model", help="path to output face mask detector model") args = vars(ap.parse_args())

vars(ap.parse_args()) # initialize the initial learning rate, number of epochs to train for,

for, # and batch size

size INIT_LR = 1e-4 EPOCHS = 20 BS = 32

32 # grab the list of images in our dataset directory, then initialize

initialize # the list of data (i.e., images) and class images

images print("[INFO] loading images...") imagePaths = list(paths.list_images(args["dataset"])) data = [] labels = []

[] # loop over the image paths

paths for imagePath in imagePaths: # extract the class label from the filename label = imagePath.split(os.path.sep)[-2]

imagePath.split(os.path.sep)[-2]
# load the input image (224x224) and preprocess it
 image = load_img(imagePath, target_size=(224, 224))
 image = img_to_array(image)
 image = preprocess_input(image)
 # update the data and labels lists, respectively
 data.append(image)
 labels.append(label)

# convert the data and labels to NumPy arrays

arrays data = np.array(data, dtype="int32") labels = np.array(labels)

np.array(labels) # perform one-hot encoding on the labels

labels lb = LabelBinarizer() labels = lb.fit_transform(labels) labels = to_categorical(labels)

to_categorical(labels) # partition the data into training and testing splits using 75% of

of # the data for training and the remaining 25% for testing

testing (trainX, testX, trainY, testY) = train_test_split(data, labels, test_size=0.25, stratify=labels, random_state=42)

random_state=42) # construct the training image generator for data augmentation

augmentation aug = ImageDataGenerator( rotation_range=20, zoom_range=0.15, width_shift_range=0.2, height_shift_range=0.2, shear_range=0.15, horizontal_flip=True, fill_mode="nearest")

fill_mode="nearest") # load the MobileNetV2 network, ensuring the head FC layer sets are

are # left off

off baseModel = MobileNetV2(weights="imagenet", include_top=False, input_tensor=Input(shape=(224, 224, 3)))

3))) # construct the head of the model that will be placed on top of the

the # the base model

model headModel = baseModel.output headModel = AveragePooling2D(pool_size=(7, 7))(headModel) headModel = Flatten(name="flatten")(headModel) headModel = Dense(128, activation="relu")(headModel) headModel = Dropout(0.5)(headModel) headModel = Dense(2, activation="softmax")(headModel)

activation="softmax")(headModel) # place the head FC model on top of the base model (this will become

become # the actual model we will train)

train) model = Model(inputs=baseModel.input, outputs=headModel)

outputs=headModel) # loop over all layers in the base model and freeze them so they will

not will # *not* be updated during the first training process

process for layer in baseModel.layers: layer.trainable = False

False # compile our model

model print("[INFO] compiling model...") opt = Adam(lr=INIT_LR, decay=INIT_LR / EPOCHS) model.compile(loss="binary_crossentropy", optimizer=opt, metrics=["accuracy"])

metrics=["accuracy"]) # train the head of the network

network print("[INFO] training head...") H = model.fit( aug.flow(trainX, trainY, batch_size=BS), steps_per_epoch=len(trainX) // BS, validation_data=(testX, testY), validation_steps=len(testX) // BS, epochs=EPOCHS)

epochs=EPOCHS) # make predictions on the testing set

set print("[INFO] evaluating network...") predIdxs = model.predict(testX, batch_size=BS)

batch_size=BS) # for each image in the testing set we need to find the index of the

the # label with corresponding largest predicted probability

probability predIdxs = np.argmax(predIdxs, axis=1)

axis=1) # show a nicely formatted classification report

report print(classification_report(testY.argmax(axis=1), predIdxs, target_names=lb.classes_))

target_names=lb.classes_)) # serialize the model to disk

disk print("[INFO] saving mask detector model...") model.save(args["model"], save_format="h5")

save_format="h5") # plot the training loss and accuracy

accuracy N = EPOCHS plt.style.use("ggplot") plt.figure() plt.plot(np.arange(0, N), H.history["loss"], label="train_loss") plt.plot(np.arange(0, N), H.history["val_loss"], label="val_loss") plt.plot(np.arange(0, N), H.history["accuracy"], label="train_acc") plt.plot(np.arange(0, N), H.history["val_accuracy"], label="val_acc") plt.title("Training Loss and Accuracy on Spectacles, Sunglasses and No Spectacles") plt.xlabel("Epoch #") plt.ylabel("Loss/Accuracy") plt.legend(loc="lower left") plt.savefig(args["plot"])