def main():
"""Download and serialize training data."""
args = create_parser().parse_args()
naip_state, naip_year = args.naip_path
download_and_serialize(
args.number_of_naips, args.randomize_naips, naip_state, naip_year,
args.extract_type, args.bands, args.tile_size,
args.pixels_to_fatten_roads, args.label_data_files, args.tile_overlap)
def main():
"""Download and serialize training data."""
args = create_parser().parse_args()
naip_state, naip_year = args.naip_path
download_and_serialize(args.number_of_naips, args.randomize_naips,
naip_state, naip_year, args.extract_type,
args.bands, args.tile_size,
args.pixels_to_fatten_roads, args.label_data_files,
args.tile_overlap)
def main():
"""Analyze each state and publish results to deeposm.org."""
naip_year = 2013
naip_states = {'de': ['http://download.geofabrik.de/north-america/us/delaware-latest.osm.pbf'],
'ia': ['http://download.geofabrik.de/north-america/us/iowa-latest.osm.pbf'],
'me': ['http://download.geofabrik.de/north-america/us/maine-latest.osm.pbf']
}
number_of_naips = 10
extract_type = 'highway'
bands = [1, 1, 1, 1]
tile_size = 64
pixels_to_fatten_roads = 3
tile_overlap = 1
neural_net = 'two_layer_relu_conv'
number_of_epochs = 10
randomize_naips = False
for state in naip_states:
filenames = naip_states[state]
raster_data_paths = download_and_serialize(number_of_naips,
randomize_naips,
state,
naip_year,
extract_type,
bands,
tile_size,
pixels_to_fatten_roads,
filenames,
tile_overlap)
model = train_on_cached_data(neural_net, number_of_epochs)
with open(CACHE_PATH + METADATA_PATH, 'r') as infile:
training_info = pickle.load(infile)
post_findings_to_s3(raster_data_paths, model, training_info, training_info['bands'], False)
requests.get('http://www.deeposm.org/refresh_findings/')
def main():
"""Analyze each state and publish results to deeposm.org."""
naip_year = 2013
naip_states = {
'de': [
'http://download.geofabrik.de/north-america/us/delaware-latest.osm.pbf'
],
'ia':
['http://download.geofabrik.de/north-america/us/iowa-latest.osm.pbf'],
'me':
['http://download.geofabrik.de/north-america/us/maine-latest.osm.pbf']
}
number_of_naips = 175
extract_type = 'highway'
bands = [1, 1, 1, 1]
tile_size = 64
pixels_to_fatten_roads = 3
tile_overlap = 1
naip_extent = None # WMIV 3/31/17
neural_net = 'two_layer_relu_conv'
number_of_epochs = 10
randomize_naips = False
for state in naip_states:
filenames = naip_states[state]
raster_data_paths = download_and_serialize(
number_of_naips, randomize_naips, state, naip_year, naip_extent,
extract_type, bands, tile_size, pixels_to_fatten_roads, filenames,
tile_overlap)
model = train_on_cached_data(neural_net, number_of_epochs)
with open(METADATA_FILE, 'r') as infile:
training_info = pickle.load(infile)
post_findings_to_s3(raster_data_paths, model, training_info,
training_info['bands'], False)
requests.get('http://www.deeposm.org/refresh_findings/')