classifier.py

"""
## Import Libraries
"""


# Common imports
import os
import numpy as np

# Visualization
import matplotlib as mpl
import matplotlib.pyplot as plt
import seaborn as sns



"""

"""


# TensorFlow imports
# may differs from version to versions

import tensorflow as tf
from tensorflow import keras

from tensorflow.keras.callbacks import ModelCheckpoint
from tensorflow.keras.callbacks import EarlyStopping

from tensorflow.keras.preprocessing import image


"""
## Splitting images
"""


# import splitfolders  # or import split_folders

# Split dengan ratio.
# Untuk hanya membagi menjadi set pelatihan dan validasi, setel tuple menjadi `ratio`, i.e, `(.8, .2)`.
# splitfolders.ratio("dataset", output="face_dataset_test_images", seed=1337, ratio=(.8,.2), group_prefix=None)


# Dataset information

# Test dataset is set explicitly, because the amount of data is very small
train_image_folder = os.path.join('datasets', 'face_dataset_train_aug_images')
test_image_folder = os.path.join('datasets', 'face_dataset_test_images')
img_height, img_width = 250, 250  # size of images
num_classes = 3 

# Training settings
validation_ratio = 0.15  # 15% for the validation
batch_size = 16

AUTOTUNE = tf.data.AUTOTUNE


# Train and validation sets
train_ds = keras.preprocessing.image_dataset_from_directory(
    train_image_folder,
    validation_split=validation_ratio,
    subset="training",
    seed=42,
    image_size=(img_height, img_width),
    label_mode='categorical',
    batch_size=batch_size,
    shuffle=True)

val_ds = keras.preprocessing.image_dataset_from_directory(
    train_image_folder,
    validation_split=validation_ratio,
    subset="validation",
    seed=42,
    image_size=(img_height, img_width),
    batch_size=batch_size,
    label_mode='categorical',
    shuffle=True)


# Test set
test_ds = keras.preprocessing.image_dataset_from_directory(
    test_image_folder,
    image_size=(img_height, img_width),
    label_mode='categorical',
    shuffle=False)


class_names = test_ds.class_names
class_names


"""
## Modelling with ResNet50
"""


base_model = keras.applications.ResNet50(weights='imagenet',
                                         include_top=False,  # without dense part of the network
                                         input_shape=(img_height, img_width, 3))

# Set layers to non-trainable
for layer in base_model.layers:
    layer.trainable = False

# Add custom layers on top of ResNet
global_avg_pooling = keras.layers.GlobalAveragePooling2D()(base_model.output)
output = keras.layers.Dense(num_classes, activation='sigmoid')(global_avg_pooling)

face_classifier = keras.models.Model(inputs=base_model.input,
                                     outputs=output,
                                     name='ResNet50')
face_classifier.summary()


# ModelCheckpoint to save model in case of interrupting the learning process
checkpoint = ModelCheckpoint("models/face_classifier.h5",
                             monitor="val_loss",
                             mode="min",
                             save_best_only=True,
                             verbose=1)

# EarlyStopping to find best model with a large number of epochs
earlystop = EarlyStopping(monitor='val_loss',
                          restore_best_weights=True,
                          patience=3,  # number of epochs with no improvement after which training will be stopped
                          verbose=1)

callbacks = [earlystop, checkpoint]


face_classifier.compile(loss='categorical_crossentropy',
                        optimizer=keras.optimizers.Adam(learning_rate=0.01),
                        metrics=['accuracy'])


"""
## Training
"""


epochs = 5

history = face_classifier.fit(
    train_ds,
    epochs=epochs,
    callbacks=callbacks,
    validation_data=val_ds)

face_classifier.save("models/face_classifier.h5")