TODO.goals

# America/New_York 8/22/2015
Consider integrating save_geoimage_from_pixel_frame
Consider implementing image.projected_frame or image.projected_bounds
Consider implementing image.select_pixel_xys
Consider implementing image.select_projected_xys
Normalize images in train_classifier
Add source_marker_folder to get_marker_from_batches
Resume automatically if the target folder already has what is needed
    Check that each section has what is needed
Write get_probabilities_from_arrays
    Show improved progress output total: partial1 partial2
Improve development classifier
    Use cuda-convnet2
        Check whether ec2 is compatible with cuda-convnet2
    Use dropout
    Consider optimizing get_counts by searching maximums locally
    Consider adding random caching to get_next_batch
    Consider making coarse classifier to optimize speed
    Use minimum and maximum to save array
Update interface
    Let them see building examples with each classifier
Rename points to locations