py-data-augmentation

Image data augmentation for learning algorithm. This repo is highly based on takmin's c++ version.

Installation Dependencies:

• Python 2.7
• numpy
• scikit-image

You can install with pip or just clone the repo:

pip install pydaag

How to use:

Your input images should be a numpy arrary with shape (img_numbers, img_rows, img_cols, img_channels).

If input images are grayscale, then the shape should be (img_numbers, img_rows, img_cols).

Simple example:

from pydaag import pydaag

#######################################################################
#inputs:
#   images: input images
#   x_slide: Maximum slide in X direction, it's ratio of width of image.
#   y_slide: Maximum slide in Y direction, it's ratio of height of image.
#   z_rotateion: Maximum rotation around Z axis.
#   y_rotateion: Maximum rotation around Y axis.
#   x_rotateion: Maximum rotation around X axis.
#   blur_max_sigma: Maximum standard deviation of Gaussian blur.
#   noise_max_sigma: Maximum standard deviation of Gaussian noise
#######################################################################
images_ = pydaag.data_augmentation(images, x_slide=0.2, y_slide=0.2, 
                                   z_rotation=20, y_rotation=20, x_rotation=20, 
                                   blur_max_sigma=3, noise_max_sigma=20)

You can find an example in test\test.py.

Run the test file:

git clone https://github.com/taoyizhi68/py-data-augmentation.git
cd py-data-augmentation\test
python test.py

inputs:

outputs: