Aug Tool is a Python library available on PyPI that simplifies image data augmentation for machine learning tasks, compatible with TensorFlow, PyTorch, and the YOLO library.
pip install aug-toolTo install "aug-tool", you can use pip, the Python package manager. Open a terminal or command prompt and run the following command:
- Supports various image augmentation techniques suitable for real environment conditions, such as adding noise, scaling, shifting, and more.
- Provides convenient integration with popular machine learning libraries such as TensorFlow, Keras, PyTorch, etc.
- Allows augmentation of both images and their annotation files in formats such as XML, or TXT.
- Customizable augmentation parameters, including rotation angle, scaling factor, flipping direction, and more.
| Labeled data | With aug tool | |
|---|---|---|
| Example 1 |  |
 |
| Example 2 |  |
 |
| Example 3 |  |
 |
| Example 4 |  |
 |
from aug_tool import Augmentation
# Specify the input parameters
open_file_name = r"C:\Users\user.name\Desktop\datas\orginal" #Path to the data folder
save_file_name = r"C:\Users\user.name\Desktop\datas\augmented"#Path to the data folder
number_of_aug = 2 # Number of augmented data to generate
# Apply data augmentation using aug-tool
Augmentation(open_data_path=open_file_name,
save_file_name=save_file_name,
number_of_aug=number_of_aug,
x_shift=15,
y_shift=15)
# Continue with further processing or analysissrc/ #root file
└── aug_tool/ #main package
├── __init__.py
├── dataOpener/ #sub package1
│ ├── __init__.py
│ ├── dataOp.py
│ ├── imgOp.py
│ ├── annOp.py
│ ├── xmlOp.py
│ └── txtOp.py
├── dataAugmentor/ #sub package2
│ ├── __init__.py
│ ├── dataAug.py
│ ├── imgAug.py
│ ├── annAug.py
│ ├── xmlAug.py
│ └── txtAug.py
├── dataSaver/ #sub package3
│ ├── __init__.py
│ ├── dataSav.py
│ ├── imgSav.py
│ ├── annSav.pys
│ ├── xmlSav.py
│ └── txtSav.py
└── augmentation.py
class Augmentation(object):
def __init__(self, open_data_path:str, save_file_name:str, number_of_aug:int, x_shift:int, y_shift:int) -> None:
"""
This is an initialization function that takes in parameters for data augmentation and saves the
augmented data and annotations to a specified file path.
:param open_data_path: The path to the directory containing the original data to be augmented
:type open_data_path: str
:param save_file_name: The name of the folder where the augmented data will be saved
:type save_file_name: str
:param number_of_aug: The number of times the image and its corresponding annotation will be
augmented
:type number_of_aug: int
:param x_shift: The amount of horizontal shift to apply during image augmentation
:type x_shift: int
:param y_shift: y_shift is a parameter that determines the maximum number of pixels by which an
image can be shifted vertically during data augmentation
:type y_shift: int
""" The __init__ method of the augmentation class performs the initialization and processing steps. This process can be divided into three parts:
In the first part, the method creates the necessary file names and initializes the folder structure for storing the augmented images. This ensures a well-organized output for the augmented data.
# It creates a folder to save inside
self.create_dest_folder(self.target_file_name)
# All the data in the file is taken and augmented in a loop
for data_path in self.create_list_of_data(open_data_path = open_data_path):
# The image and label file with the same name as image is opened once
image = self.label_factory(data_path[:-4])
label = self.image_factory(data_path)
# The augmentation process is performed as much as the `number_of_aug` value from the user
for num in range(number_of_aug):
# The name of new file that will be augmented is created
data_name = data_path.split("\\")[-1][:-4] + "_aug" + str(num + 1))Here, The self.create_dest_folder method is responsible for creating the output folder if it doesn't already exist. self.create_list_of_data stores the path of the files which will be augmented . The loop iterates over the files in the self.create_list_of_data list and obtains the paths to the corresponding image and label files. The self.label_factory and self.image_factory methods are responsible for opening the image and label files, respectively.
The second part of the __init__ method performs the actual augmentation processing of images. Within a nested loop, the augmentation techniques are applied to the image to generate multiple augmented versions of the data.
# The augmented image is created acording values that given users
aug_image = dataAugmentor.ImgAug(image=image,
x_shift=x_shift,
y_shift=y_shift).image_aug
# The augmented image is saved on target file
dataSaver.ImgSav(target_file_path = self.target_file_name,
img_aug = aug_image,
data_name=data_name)The third part of the __init__ method performs the actual augmentation processing of labels. Within a nested loop, the augmentation techniques are applied to the label data to generate multiple augmented versions of the data.
# If the extension of the read label file is '.xml', it is processed here
if self.ann.ext == ".xml":
ann_aug = dataAugmentor.XmlAug(name=data_name,
annotate= label.data,
x_shift=x_shift,
y_shift=y_shift)
dataSaver.XmlSav(target_file_path = self.target_file_name,
ann_aug=ann_aug.annotate,
data_name=data_name)
# If the extension of the read label file is '.txt', it is processed here
elif self.ann.ext == ".txt":
ann_aug = dataAugmentor.TxtAug(
annotate= label.data,
x_shift=x_shift,
y_shift=y_shift,
width=aug_image.width,
height=aug_image.height).aug_anotate
dataSaver.TxtSav(target_file_path = self.target_file_name,
ann_aug=ann_aug,
data_name=data_name)This project is licensed under the MIT License. You are free to use, modify, and distribute this library in accordance with the terms specified in the license.
If you have any questions, suggestions, or need support, feel free to reach out to [email protected]