ML-keras-hackerearth-WBC-Segmentaion

Code for WBC Segementation, a hackerearth compeition held by sigtuple

Requirements:

• python 3.5
• anaconda 4.2.0
• tensorflow - conda
• keras
• theano

To Run:

• Put the train data and test data folder in the Data folder
• First run the data.py file to load the data and preprocess it.
• Then run the train.py file to train a Convolutional Neural Network, make prediction and store the predicted images.
• The predicted mask images can be found in the Data/output folder.

Approach:

• The images are converted into arrays and stored.
• A Convolutional Neural Network(CNN) having 5 convolution2d block of filter values 64, 32 and 1. Each layer is seperated by an activation layer(Relu).
• The CNN is trained with the training data and then made to predict the output(mask images) for the testing data.
• The output is saved back as an image in the Data/output folder.