AUG-alpha

[Warning, Unmaintained!] Alternatives: imgaug, albumentations, kornia.

AUG is an open source augmentation library based on OpenCV. The repository contains a set of image transformations for data augmentation and artificial data synthesis.

Major features:

• supports operations on masks and points,
• unified interface - most magnitude coefficients are in the range [0, 1],
• optimized operations,
• unique operations compared to imgaug/albumentations,
• Python 3, OpenCV 4.1.

Installation

Pip:

pip install aug

The latest version, directly from github:

pip install -U git+https://github.com/cta-ai/aug

Example operations

Jitter	Radial gradient	Channel Shuffle	Cutout
Blend with random images	CLAHE	Contrast	Zoom
Salt and pepper	Dilation	Erosion	Texture modification
Flashlight	Flips	Gamma	Random shadow
Gaussian noise	Brightness	Inversion	Rotation90
Gaussian blur	Motion blur	Variable blur	Pixelize
Median blur	Linear gradient	JPEG noise	Random Curve
Elastic transformation	Optical transformation	Perspective transformation	Grid distortion
Rotation	Scratches	Halo effect

Example usage:

import aug

class SimpleExample(aug.Pipeline):
    def __init__(self):
        super(SimpleExample, self).__init__()
        self.seq = aug.Sequential(
            aug.Rotation(p=.5, angle=90),
            aug.GaussianBlur(p=1.),
        )

    def apply(self, sample):
        return self.seq.apply(sample)
        

sample = SimpleExample().apply(aug.Sample(image, annotations, masks))

More: Getting started.

Releases

v0.1.0 - 16/07/2019

• Initial alpha release.

Licence

Apache License 2.0

Contact

Project is maintained mainly by Tomasz Gilewicz (@tgilewicz).