Automatic photo-from-panorama for Google Maps

We introduce a technique for extracting interesting photographs from 360° panoramas. We build on the success of convolutional neural networks for classification to train a model that scores a given view, using this score to find a best view. Training data for this classification model is generated automatically from landmark detections within Street View panoramas. We validate that our selected views are often preferred over manually chosen ones and have experienced an increase in user interaction when automatically selected views are shown on Google Maps.