Using TensorFlow to classify aerial views of images using deep learning.

The Last Layer of the Inception Model by Google is trained on hundreds of images of aerial views of major US Cities.

Seattle

San-Francisco

New-York

Chicago

After creating bottlenecks of all the images in each city folder, I obtained a test data accuracy of 56.4%

Upon feeding novel images of each city to the network, the network predicted the correct city 3 out of 4 times. (EXCITING!!!)

The network predicted Seattle incorrectly (~ 14.5% confidence)

The network predicted San-Francisco correctly (~91% confidence)

The network predicted New-York correctly (~88% confidence)

The network predicted Chicago correctly (~91% confidence)

Here are the raw results.