Machine Learning Fall 2019 Project 1
Humans are exceptional at determining location by using cues like driving direction, vegetation, weather, languages, and other less evident facets. We attempt to train a deep neural network that uses the same human reasoning provided Google street view images to determine what state in the U.S. the picture was taken.
The dataset we are using is a collection of Google street view images found on DeepGeo: Photo Localization with Deep Neural Network. Each state contains 10K images divided into a training(90%) and test(10%) set. Each input sample contains 4 images facing each cardinal direction.
In the works! Check back in a week :)
How can we capture the effects of factors that are not binary (like architecture styles or reading signboards)? Do we have to create binary layers that add up to these multi-faceted features?