augmenter 0.1.0

A toolkit for automated data augmentation

Augmenter: Advanced Image Augmentation Library

Augmenter is a Python library designed to simplify image augmentation for machine learning and deep learning projects. It provides an easy-to-use interface to apply a variety of augmentation techniques and supports randomized augmentation mixing to generate diverse variations of each image.

---

Features

• Augment Entire Datasets: Process all images in a directory, including nested subdirectories, while preserving the folder structure.
• Extensive Augmentation Techniques: Includes operations like flipping, rotation, blurring, brightness/contrast adjustment, grayscale conversion, noise addition, sharpening, and cropping.
• Random Augmentation Mixing: Combine random augmentations to create diverse variations of the same image.
• Customizable Variations: Control the number of augmented variations per image when using mixing.
• Structured Output: The augmented images are saved in the same directory structure as the original dataset.

---

Installation

Install Augmenter directly from PyPI:

pip install augmenter

---

Usage-1

Basic Augmentation

To apply augmentations to all images in a folder, preserving the directory structure, use the following code:

from augment_labeler import Augmenter

augmenter = Augmenter()

# Applying augmentations with mixing enabled and 3 variations per image
augmented_count = augmenter.apply(
    image_dir="./dataset",  # Path to the dataset
    augmentations=[
        "flip", 
        "rotate", 
        "blur", 
        "brightness_contrast", 
        "grayscale", 
        "noise", 
        "sharpen", 
        "crop"
    ],  # List of augmentations to apply
    save_path="./augmented_dataset",  # Path to save augmented images
    mixing=True,  # Enable mixed augmentations
    variations_per_image=3  # Create 3 variations for each image
)

print(f"Total augmented images created: {augmented_count}")

Output Structure

After running the above script, the output will be saved in the augmented_dataset folder, preserving the original folder structure with the added augmentation variations. The directory structure will look like this:

augmented_dataset/
├── class1/
│   ├── img1_mix_1.jpg
│   ├── img1_mix_2.jpg
│   ├── img1_mix_3.jpg
│   ├── img2_mix_1.jpg
│   ├── img2_mix_2.jpg
│   ├── img2_mix_3.jpg
├── class2/
│   ├── img3_mix_1.jpg
│   ├── img3_mix_2.jpg
│   ├── img3_mix_3.jpg
│   ├── img4_mix_1.jpg
│   ├── img4_mix_2.jpg
│   ├── img4_mix_3.jpg

• Each image is augmented and saved with a _mix_X suffix (where X is the variation number).
• The folder structure is preserved for each class (e.g., class1/, class2/), making it easy to use the augmented dataset for machine learning.

---

Usage-2

Basic Augmentation

To apply augmentations to all images in a folder, preserving the directory structure, use the following code:

from augment_labeler import Augmenter

augmenter = Augmenter()
augmented_count = augmenter.apply(
    image_dir="./dataset", 
    augmentations=["flip", "rotate", "blur"], 
    save_path="./augmented_dataset"
)

print(f"Total augmented images created: {augmented_count}")

Output Structure

After running the above script, the output will be saved in the augmented_dataset folder, preserving the original folder structure with the added augmentation variations. The directory structure will look like this:

/your_project/
├── dataset/
│   ├── class1/
│   │   ├── img1.jpg
│   │   ├── img2.jpg
│   ├── class2/
│   │   ├── img3.jpg
│   ├── img4.jpg
├── augmented_dataset/
│   ├── class1/
│   │   ├── img1_flip.jpg
│   │   ├── img1_rotate.jpg
│   │   ├── img1_blur.jpg
│   │   ├── img2_flip.jpg
│   │   ├── img2_rotate.jpg
│   │   ├── img2_blur.jpg
│   ├── class2/
│   │   ├── img3_flip.jpg
│   │   ├── img3_rotate.jpg
│   │   ├── img3_blur.jpg
│   ├── img4_flip.jpg
│   ├── img4_rotate.jpg
│   ├── img4_blur.jpg

• The folder structure is preserved for each class (e.g., class1/, class2/), making it easy to use the augmented dataset for machine learning.

---

Supported Augmentations

• flip: Horizontal flip
• rotate: Random rotation (up to ±45 degrees)
• blur: Gaussian blur
• brightness_contrast: Random brightness and contrast adjustments
• grayscale: Convert image to grayscale
• noise: Add Gaussian noise
• sharpen: Sharpen the image
• crop: Random resized cropping

---

---

Contributing

Contributions are welcome! To contribute:

1. Fork the repository.
2. Create a new branch for your feature or bugfix.
3. Commit your changes and push them to your fork.
4. Open a pull request.

---

License

This project is licensed under the MIT License. See the LICENSE file for details.

---

Release Notes

v0.1.0

• Initial release.
• Support for multiple augmentations and structured dataset output.
• Added random augmentation mixing and customizable variations.

---

Author

Developed by AKM Korishee Apurbo. Feel free to reach out with questions, suggestions, or contributions!