Process, Augment, and Balance Image Data
This library is designed to facilitate the preprocessing phase of image classification projects in order to get into the fun part: training the models!
Imgo is composed of two modules: uptools and augtools.
Uptools helps to streamline various image data preprocessing tasks, such as:
- Reading images from a local disk
- Rescaling images
- Normalizing and standardizing pixel values
- Converting image datasets into numpy-arrays
- One-hot-encoding label data
- Splitting image datasets into training, validation, and testing subsets
- Merging data subsets into a single dataset
- Saving numpy-arrays as images in class subdirectories
Augtools allows the user to quickly and efficiently apply augmentation to image data. With Augtools, users can perform the following augmentation tasks using very few lines of code:
- Apply a powerful collection of transformation and corruption functions
- Augment images saved on a local disk
- Save augmented images in class subdirectories
- Augment entire image datasets
- Augment training data in place in preparation for machine learning projects
- Rebalance class sizes by generating new training images
It's as easy as
pip install imgo!
Have a look at the demos here!
Documentation is currently available in the form of docstrings.
Requirements and Dependencies
Please see the
requirements.txt file for all requirements and dependencies.
Issues / To do
Some functions currently employ ragged arrays which are deprecated in the latest versions of NumPy. This affects all functions that work with non-standard or inconsistent image dimensions.
The project is licensed under the MIT license.
Some of the augtools library is built as a wrapper around Imgaug, a powerful image augmentation library. For more information, please see https://imgaug.readthedocs.io/en/latest/.