This is a simple example using the Galaxy Zoo dataset (also seen in the Galaxy Challenge on kaggle). This example also demonstrates how a large dataset can be imported into DeepForge and streamed to the neural network during training.

The example project contains the training data and can be downloaded from here. Or, if you prefer downloading from the console:

wget -O galaxy-data.webgmex https://deepforge.s3.amazonaws.com/galaxy-data.webgmex?AWSAccessKeyId=AKIAISTYNTKQZLV6IVLQ&Expires=1633932738&Signature=8%2BlF1x%2FsF0fGE0K0DGBLjG1R9v0%3D

In this example, we are simply cropping a 106x106 section of the image and training either a feed-forward neural network or a convolutional neural network. The architecture for the convolutional neural network is based on the paper by Sandor Dieleman. The entire citation for the paper is given below.

@article{dieleman2015rotation,
  title={Rotation-invariant convolutional neural networks for galaxy morphology prediction},
  author={Dieleman, Sander and Willett, Kyle W and Dambre, Joni},
  journal={Monthly notices of the royal astronomical society},
  volume={450},
  number={2},
  pages={1441--1459},
  year={2015},
  publisher={Oxford University Press}
}

This example depends on the deepforge-keras extension.