Tensorflow solution for the Dark Skies Challenge
This tutorial focuses on Dark Skies - Classification of Nighttime Images, an interesting challenge on crowdAI.
The goal of this challenge is to use the manually labeled dataset to develop an image classification algorithm that can correctly identify whether an photo shows stars, cities, or other objects. These photos were taken at night. There are 101,554 images in the training set and 52,317 images in the test set.
The Iberian Peninsula at night, showing Spain and Portugal. Madrid is the bright spot just above the center. Credits: NASA
This tutorial is structured as a hand-on document. Details about machine learning, neural networks or convolution neural networks are not discussed.
If you are new to machine learning, I recommend you follow this excellent course by Andrew Ng from Cousera: https://www.coursera.org/learn/machine-learning. In case you are already familiar with machine learning, the CS231n course from Standford (http://cs231n.stanford.edu) will be a great resource when dealing with visual recognition problem.
This tutorial also uses Convolution Neural Network (CNN), which described extensively in CS231n course. The model is based on Google Inception v3 architecture on Tensorflow framework. Other scripts are written in Python.