Random Pixel Drop

This is a library written for ImageDataGenerator() from keras.preprocessing library.

Random pixel drop is inspired from dropout regularisation technique and cutout image augmentation technique. This library will randomly set image pixel values of the image to zeros for a given range of number. Which will be similar to dropout than switching neurons randomly to reduce over fitting this will introduce randomness in the input image to reduce over fitting

Implementation

git clone https://github.com/chekoduadarsh/Random-Pixel-Drop

The above given command will download the repository to your working directory, then in your python code tou can import the library as following,

With ImageDataGenerator in Keras

It is very easy to use if you are using ImageDataGenerator in Keras, get drop function from get_random_pixel_drop() which can be passed to ImageDataGenerator as preprocessing_function. This function uses 3 hyper parameters

Drop_num -> Number of dropped pixels

l_rand_bound -> lower boundry for random pixel value generation

u_rand_bound -> upper boundry for random pixel value generation