Skip to main content

A tiny python library to augment the images dataset aimed for a ML classification system

Project description

Table of Contents


The scope of this library is to augment the dataset for an image classification ML system.



The library is compatible with python 3.6 on.


We suggest to isolate your installation via python virtualenv:

python3 -m venv .imgaug
source .imgaug/bin/activate


Update pip package manager:

pip install pip --upgrade
pip install -r requirements.txt


The library is covered, by fast, isolated unit and doc testing (the latter to grant reliable documentation):

python -m unittest discover -s imgaug -p '*'


The library is composed by different collaborators, each with its specific responsibility. Each class tries to expose a minimal public APIs in the form of __call__ or __iter__ methods (when generators are used).
The classes are aimed to work with one image at time, in case you need to transform and augment multiple images, avoid creating multiple instances of the classes, just change the argument of the __call__ function (but for Persister, which need a new instance and/or instance attribute modification).


The target label is extracted directly by inspecting the image name and trying to extract meaningful information (customisable).

lbl = Labeller(digits=10)




The images are normalized by:

  • resizing them to the specified max size (default to 256 pixels)
  • optionally applying a squared, transparent/backgound canvas and centering the image on it, thus avoiding any deformation
norm = Normalizer(size=128, canvas=True)
img = norm('resources/bag.png')
(128, 128, 4)


The number of images is augmented by two orders of magnitude (depending on the cutoff float attribute) by applying different transformations to the original one.
Transformations are applied by using generators, thus saving memory consumption.

aug = Augmenter(cutoff=.5)
<generator object Augmenter.__call__ at 0x125354480>


Images are persisted upon normalization and augmentation, by specifying an action function that accepts the name of the file (original basename suffixed by an index) and a BytesIO object containing the image data stream.
The persister supports both a filename path and, optionally, a stream-like object (in case the file is not yet persisted to disk).
The persister supports iteration by yielding the image label and the function return value (typically the saved path), allowing to generate CSV files specific to cloud platforms (i.e. Google Vision APIs).

def persist(name, stream):
    filename = f'temp/{name}'
    with open(filename, 'wb') as f:
    return filename

pers = Persister('resources/skirt.jpg', action=perist)
for label, filename in pers:
    print(label, filename)


In case you need an archive with each normalised augmentations within the recognised label subfolder, you can rely on the Zipper interface: it creates a ZIP file on current path, by scanning the specified folder for PNG or JPG images.

zipper = Zipper('.resources/', normalizer=image.Normalizer(16), augmenter=image.Augmenter(.05))

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for image-augmenter, version 0.5.5
Filename, size File type Python version Upload date Hashes
Filename, size image_augmenter-0.5.5-py3-none-any.whl (11.9 kB) File type Wheel Python version py3 Upload date Hashes View
Filename, size image_augmenter-0.5.5.tar.gz (8.9 kB) File type Source Python version None Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page