SpectraFaceAuth/AttendanceApp

1import numpy as np
2import pandas as pd
3import csv
4import tensorflow as tf
5from sklearn.model_selection import train_test_split
6import cv2
7from pathlib import Path
8from tensorflow.keras.models import Sequential
9from tensorflow.keras.layers import Dense, Flatten, Input
10from tensorflow.keras.optimizers import Adam
11from keras.applications import vgg16
12
13def ModelFineTuning():
14    # Define the path to your dataset
15    data_dir = Path('Dataset')
16    image_size = (224, 224)  # VGGFace model expects 224x224 images
17
18    # Initialize dictionaries
19    candidates_dict = {}
20    labels_dict = {}
21
22    # Get all class folder names
23    class_folders = [folder.name for folder in data_dir.iterdir() if folder.is_dir()]
24    total_classes = len(class_folders)
25
26    # Assign labels to each class
27    for idx, class_name in enumerate(class_folders):
28        candidates_dict[class_name] = list(data_dir.glob(f'{class_name}/*'))
29        labels_dict[class_name] = idx
30
31    df = pd.DataFrame(list(labels_dict.items()), columns=['Candidate Name', 'Label'])
32    df.to_csv("candidate_labels.csv", index=False)
33
34    # Print the results
35    print('Images Dictionary:')
36    print(candidates_dict)
37    print('\nLabels Dictionary:')
38    print(labels_dict)
39
40    X, y = [], []
41    if len(candidates_dict.items()) == 0:
42        return False
43    for candidate_name, faces in candidates_dict.items():
44        for image in faces:
45            img = cv2.imread(str(image))
46            resized_img = cv2.resize(img, image_size)
47            X.append(resized_img)
48            y.append(labels_dict[candidate_name])
49
50    print(len(X))
51
52    X = np.array(X)
53    y = np.array(y)
54
55    X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0)
56
57    X_train_scaled = X_train / 255.0
58    X_test_scaled = X_test / 255.0
59
60    # Convert labels to one-hot encoding
61    y_train = tf.keras.utils.to_categorical(y_train, num_classes=total_classes)
62    y_test = tf.keras.utils.to_categorical(y_test, num_classes=total_classes)
63
64    # Load the pre-trained VGGFace model
65    base_model = vgg16.VGG16(weights='imagenet', include_top=False, input_shape=(224, 224, 3))
66
67    # Ensure the base model layers are not trainable
68    for layer in base_model.layers:
69        layer.trainable = False
70
71    # Create a Sequential model and add layers
72    model = Sequential()
73    model.add(Input(shape=(224, 224, 3)))
74    model.add(base_model)
75    model.add(Flatten())
76    model.add(Dense(1024, activation='relu'))
77    model.add(Dense(512, activation='relu'))
78    model.add(Dense(total_classes, activation='softmax'))
79
80    # Compile the model
81    model.compile(optimizer=Adam(learning_rate=0.0001), loss='categorical_crossentropy', metrics=['accuracy'])
82
83    # Train the model
84    history = model.fit(
85        X_train_scaled, y_train,
86        validation_data=(X_test_scaled, y_test),
87        epochs=10,  # Adjust the number of epochs based on your needs
88        batch_size=32
89    )
90
91    # Evaluate the model
92    loss, accuracy = model.evaluate(X_test_scaled, y_test)
93    print(f"Test accuracy: {accuracy * 100:.2f}%")
94
95    # Save the fine-tuned model
96    model.save('fine_tuned_VGG16_model.h5')
97    return True
98
99# ModelFineTuning()  # Uncomment this line to run the training
100