fagan2888 / data-augmentation

These functions will randomly distort image data to amplify total amount of images for use in machine learning algorithms.

Data-Augmentation

These functions will randomly distort image data to amplify total amount of images for use in machine learning algorithms.

We will use this package to amplify ("augment") a single image.

The library randomly rotates, flips along axis, translates, adds noise, and adds "black boxes" to images to increase the number of usable data for a machine learning algorithm.

import DataAugmentation as da

import matplotlib.pyplot as plt
import numpy as np
%matplotlib inline

Augment function signature:
def augment(images, labels=None, amplify=2):

Input Data:

images shape: (batch_size, height, width, channels=3)
labels shape: (batch_size, 3) [this is an optional parameter]
amplify is how many (randomly altered) copies of the original image(s) you want to produce

# Sample image to amplify
sample = plt.imread('samples/dog_icon.jpg')
plt.imshow(sample)

<matplotlib.image.AxesImage at 0x1161fe550>

sample.shape #Need to reshape to (batch_size, height, width, channels)
sample = np.reshape(sample, (1,256,256,3))

augmented_data = da.augment(sample, amplify=10)

augmented_data.shape
#First image is original, rest are synthesized

(10, 256, 256, 3)

plt.imshow(augmented_data[1])
#Randomly translated

<matplotlib.image.AxesImage at 0x118a3bd68>

plt.imshow(augmented_data[7])
#Look carefully, random noise added

<matplotlib.image.AxesImage at 0x118e4e080>

plt.imshow(augmented_data[6])
#Randomly rotated

<matplotlib.image.AxesImage at 0x118f488d0>