File size: 9,785 Bytes
ffd6b68
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
import tensorflow as tf
import os
import argparse
from sklearn.model_selection import StratifiedShuffleSplit
from tqdm import tqdm
import uuid
import random

# Parses command line arguments
def parse_arguments():
    parser = argparse.ArgumentParser(description='Image Data Loader with Augmentation and Splits')
    parser.add_argument('--path', type=str, required=True, help='Path to the folder containing images')
    parser.add_argument('--dim', type=int, default=224, help='Required image dimension')
    parser.add_argument('--batch_size', type=int, default=32, help='Batch size')
    parser.add_argument('--target_folder', type=str, required=True, help='Folder to store the train, test, and val splits')
    parser.add_argument('--augment_data', action='store_true', help='Apply data augmentation')
    parser.add_argument('--balance', action='store_true', help='Balance the dataset')
    parser.add_argument('--split_type', type=str, choices=['random', 'stratified'], default='random',
                        help='Type of data split (random or stratified)')
    return parser.parse_args()

# Process the input images
def process_image(file_path, image_size):
    image = tf.io.read_file(file_path)
    image = tf.image.decode_image(image, channels=3, dtype=tf.float32)
    image = tf.image.resize(image, [image_size, image_size])
    image = tf.clip_by_value(image, 0.0, 1.0)
    return image

# Balances the images of a specific class
def balance_class_images(image_paths, labels, target_count, image_size, label, label_to_index, output_folder):
    print(f"Balancing class '{label}'...")
    label_idx = label_to_index.get(label, None)
    if label_idx is None:
        print(f"Label '{label}' not found in label_to_index.")
        return [], []

    image_paths = [img for img, lbl in zip(image_paths, labels) if lbl == label_idx]
    num_images = len(image_paths)

    print(f"Class '{label}' has {num_images} images before balancing.")

    balanced_images = []
    balanced_labels = []

    original_count = num_images
    synthetic_count = 0

    if num_images > target_count:
        balanced_images.extend(random.sample(image_paths, target_count))
        balanced_labels.extend([label_idx] * target_count)
        print(f"Removed {num_images - target_count} images from class '{label}'.")
    elif num_images < target_count:
        balanced_images.extend(image_paths)
        balanced_labels.extend([label_idx] * num_images)

        num_to_add = target_count - num_images
        print(f"Class '{label}' needs {num_to_add} additional images for balancing.")

        while num_to_add > 0:
            img_path = random.choice(image_paths)
            image = process_image(img_path, image_size)

            for _ in range(min(num_to_add, 5)):  # Use up to 5 augmentations per image
                augmented_image = augment_image(image)
                balanced_images.append(augmented_image)
                balanced_labels.append(label_idx)
                num_to_add -= 1
                synthetic_count += 1

        print(f"Added {synthetic_count} augmented images to class '{label}'.")
        print(f"Class '{label}' has {len(balanced_images)} images after balancing.")

    class_folder = os.path.join(output_folder, str(label_idx))
    if not os.path.exists(class_folder):
        os.makedirs(class_folder)

    for i, img in enumerate(balanced_images):
        file_name = f"{uuid.uuid4()}.png"
        file_path = os.path.join(class_folder, file_name)
        save_image(img, file_path)

    print(f"Saved {len(balanced_images)} images for class '{label}' (Original: {original_count}, Synthetic: {synthetic_count}).")

    return balanced_images, balanced_labels

# Saves an image to a file
def save_image(image, file_path):
    if isinstance(image, str):
        image = process_image(image, image_size)
    if isinstance(image, tf.Tensor):
        image = tf.image.convert_image_dtype(image, dtype=tf.uint8)
        image = tf.image.encode_png(image)
    else:
        raise ValueError("Expected image to be a TensorFlow tensor, but got a different type.")

    tf.io.write_file(file_path, image)

# Augments an image with random transformations
def augment_image(image):
    # Apply random augmentations using TensorFlow functions
    image = tf.image.random_flip_left_right(image)
    image = tf.image.random_flip_up_down(image)
    image = tf.image.random_brightness(image, max_delta=0.1)
    image = tf.image.random_contrast(image, lower=0.9, upper=1.1)
    image = tf.image.random_saturation(image, lower=0.9, upper=1.1)
    image = tf.image.random_hue(image, max_delta=0.1)
    return image

# Creates a list of data augmentation functions
def create_datagens():
    return [augment_image]

# Balances the entire dataset by balancing each class
def balance_data(images, labels, target_count, image_size, unique_labels, label_to_index, output_folder):
    print(f"Balancing data: Target count per class = {target_count}")

    all_balanced_images = []
    all_balanced_labels = []

    for label in tqdm(unique_labels, desc="Balancing classes"):
        num_images = len([img for img, lbl in zip(images, labels) if lbl == label_to_index.get(label, -1)])
        balanced_images, balanced_labels = balance_class_images(
            images, labels, target_count, image_size, label, label_to_index, output_folder
        )
        all_balanced_images.extend(balanced_images)
        all_balanced_labels.extend(balanced_labels)

    total_original_images = sum(1 for img in all_balanced_images if isinstance(img, str))
    total_synthetic_images = len(all_balanced_images) - total_original_images

    print(f"\nTotal saved images: {len(all_balanced_images)} (Original: {total_original_images}, Synthetic: {total_synthetic_images})")

    return all_balanced_images, all_balanced_labels

# Augments an image using TensorFlow functions
def tf_augment_image(file_path, label):
    image = tf.image.resize(tf.image.decode_jpeg(tf.io.read_file(file_path)), [image_size, image_size])
    image = tf.cast(image, tf.float32) / 255.0
    augmented_image = augment_image(image)
    return augmented_image, label


def map_fn(file_path, label):
    image, label = tf.py_function(tf_augment_image, [file_path, label], [tf.float32, tf.int32])
    image.set_shape([image_size, image_size, 3])
    label.set_shape([])
    return image, label

# Loads images, splits them into train, validation, and test sets, and saves the splits
def load_and_save_splits(path, image_size, batch_size, balance, datagens, target_folder, split_type):
    all_images = []
    labels = []

    for class_folder in os.listdir(path):
        class_path = os.path.join(path, class_folder)
        if os.path.isdir(class_path):
            for img_file in os.listdir(class_path):
                img_path = os.path.join(class_path, img_file)
                all_images.append(img_path)
                labels.append(class_folder)  # Use the folder name as the label

    print(f"Loaded {len(all_images)} images across {len(set(labels))} classes.")
    print(f"Labels found: {set(labels)}")  # Print unique labels

    unique_labels = list(set(labels))
    label_to_index = {label: idx for idx, label in enumerate(unique_labels)}
    encoded_labels = [label_to_index[label] for label in labels]

    print(f"Label to index mapping: {label_to_index}")

    if split_type == 'stratified':
        sss = StratifiedShuffleSplit(n_splits=1, test_size=0.2, random_state=42)
        train_indices, test_indices = next(sss.split(all_images, encoded_labels))
    else:  # random split
        total_images = len(all_images)
        indices = list(range(total_images))
        random.shuffle(indices)
        train_indices = indices[:int(0.8 * total_images)]
        test_indices = indices[int(0.8 * total_images):]

    train_files = [all_images[i] for i in train_indices]
    train_labels = [encoded_labels[i] for i in train_indices]
    test_files = [all_images[i] for i in test_indices]
    test_labels = [encoded_labels[i] for i in test_indices]

    # Create validation and test sets
    sss_val = StratifiedShuffleSplit(n_splits=1, test_size=0.5, random_state=42)
    val_indices, test_indices = next(sss_val.split(test_files, test_labels))

    val_files = [test_files[i] for i in val_indices]
    val_labels = [test_labels[i] for i in val_indices]
    test_files = [test_files[i] for i in test_indices]
    test_labels = [test_labels[i] for i in test_indices]

    # Save splits
    for split_name, file_list, labels_list in [("train", train_files, train_labels), ("val", val_files, val_labels), ("test", test_files, test_labels)]:
        split_folder = os.path.join(target_folder, split_name)
        os.makedirs(split_folder, exist_ok=True)
        with open(os.path.join(split_folder, f"{split_name}_list.txt"), 'w') as file_list_file:
            for img_path, label in zip(file_list, labels_list):
                label_folder = os.path.join(split_folder, str(label))
                if not os.path.exists(label_folder):
                    os.makedirs(label_folder)
                file_list_file.write(f"{img_path}\n")
                save_image(img_path, os.path.join(label_folder, f"{uuid.uuid4()}.png"))

    print(f"Saved splits: train: {len(train_files)}, val: {len(val_files)}, test: {len(test_files)}.")

# Main function to run the data loader
def main():
    args = parse_arguments()
    load_and_save_splits(args.path, args.dim, args.batch_size, args.balance, create_datagens(), args.target_folder, args.split_type)

if __name__ == "__main__":
    main()