image-model-augmentation-library 0.2.0

A lightweight image augmentation package

Image Augmentation Library

Created by Dylan Tran

This library is implements an image processing and augmentation pipeline for model training purposes. It has a few key features:

Geometric Augmentation

• From one image, various rotations and reflections are applied to generate additional images.
• Transformations include 90, 180, and 270 degree rotation, as well as reflection across x and y axes.

Rectangular Patch Augmentation

• This type of transformation occludes part of an original image for use with unet segmentation/prediction models. This is useful in cases where the ground truth is known, and pairs of incomplete/complete images are needed for training.
• Rectangle generation is random, meaning that the user can specify how many images can be generated from one source image.
• The color of the rectangular patch can be modified. In this case, it is white for display purposes.

Contrast Adjustment

• This image processing step increases contrast between pixels, making features more visually apparent.

Example usage:

• Note that, when possible, the pipe-and-filter style is implemented.
• This also means that various augmentations can be mixed and matched to vastly increase the training data.

imageArr = readImageDirIntoArr(r"C:\Augmentation\test_images\msk_3_7")
startIdx = 0
for augArr in rectangleCrop(imageArr, 5):
    saveOutputs(r"C:\Augmentation\sample_output", augArr, startIdx)
    startIdx += len(augArr)