skimage-portable-augmentation

Introduction

Skimage-portable-augmentation is a portable and parallel image augmentor, implemented by skimage(scikit-image) library. If you just want to make a augmentation, and give me a directory to get augmented images, this project actually is what you want.

Examples

in_path should be the directory which contain all images, and each of subdirs denotes a class;
out_path should be the output directory, and the output directory will have the same structure of subdirs with the in_path;
num_thread should be the number of threads to process the augmentation;
extension should be the extension to be used when generating new images;
verbose should be a bool value to control whether to show the aumentation info.\

if you want to put new images in the in_folder, just ignore out_folder parameter

#augmentor = ImageAugmentor("./images")
augmentor = ImageAugmentor("./images", num_thread=8, extension="jpg", out_folder="./augmentation", verbose=True)
augmentor.augmentation()

Augmentation Types

• color inversion
• exposure
• gamma
• log correlation
• random noise
• rescale
• rgb2gray
• rotate
• sigmoid correlation

Each image will get 12 new images. If you want add new type, just open the file image_augmentation.py. And add a new member function like this:

def new_type(self, image):
    return new_operation(image)

Then add one new line at the end of function single_augmentation:

self.__augmentation_and_save(self.new_type, name, img, out_folder)

It's OK.

Type	Augmented Image
original image
color inversion
exposure
gamma
log correlation
random noise
rescale
rgb2gray
rotate
sigmoid correlation

Reference

kaggle notebook