Synopsis

In this project we trained a distance function for local region matching. We show that when the application is relatively constrained, using a supervised learning approach produces better results than a generic system that was tuned by hand.