Skip to content
master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
lib
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

unmapped

Retrains inception-v3 ConvNet on labeled imagery from mapbox-satellite to automate the search for unmapped roads in OSM.

See rodowi/mapscan for a point-and-click interface to this classifier.

Rationale

missing roads

Still many roads to map in OSM.

Instructions

Generate training set

% export MAPBOX_ACCESS_TOKEN=pk.1001.foobar
% ./scripts/data.sh
% ./scripts/density.sh
% ./scripts/imagery.sh

Tiles containing highways in OSM.

map sat

Training

Still working on documenting this part

✗ docker run -it -v $HOME/c/unmapped/imagery:/tf_files/satellite gcr.io/tensorflow/tensorflow:latest-devel

root@3993bf4c0be8:~ cd /tensorflow/
root@3993bf4c0be8:/tensorflow# python tensorflow/examples/image_retraining/retrain.py --bottleneck_dir=/tf_files/bottlenecks --output_graph=/tf_files/retrained_graph.pb --output_labels=/tf_files/retrained_labels.txt --image_dir /tf_files/satellite
>> Downloading inception-2015-12-05.tgz 100.0%
Successfully downloaded inception-2015-12-05.tgz 88931400 bytes.
Looking for images in 'highway'
Looking for images in 'noway'
Creating bottleneck at /tf_files/bottlenecks/highway/11448-26515-16.jpg.txt
Creating bottleneck at /tf_files/bottlenecks/highway/11451-26502-16.jpg.txt
Creating bottleneck at /tf_files/bottlenecks/highway/11455-26498-16.jpg.txt
...
2017-01-04 05:23:14.191868: Step 0: Train accuracy = 88.0%
2017-01-04 05:23:14.192086: Step 0: Cross entropy = 0.645307
2017-01-04 05:23:14.650036: Step 0: Validation accuracy = 79.0%
2017-01-04 05:23:18.811447: Step 10: Train accuracy = 81.0%
2017-01-04 05:23:18.811610: Step 10: Cross entropy = 0.472717
2017-01-04 05:23:19.189587: Step 10: Validation accuracy = 85.0%
...
2017-01-01 03:02:53.240915: Step 490: Train accuracy = 96.0%
2017-01-01 03:02:53.241116: Step 490: Cross entropy = 0.167174
2017-01-01 03:02:53.588690: Step 490: Validation accuracy = 87.0%
2017-01-01 03:02:56.710327: Step 499: Train accuracy = 91.0%
2017-01-01 03:02:56.710478: Step 499: Cross entropy = 0.211515
2017-01-01 03:02:57.068353: Step 499: Validation accuracy = 84.0%
Final test accuracy = 88.8%

Prediction

root@5cca0bc5d586:/tensorflow# bazel-bin/tensorflow/examples/label_image/label_image --graph=/tf_files/retrained_graph.pb --labels=/tf_files/retrained_labels.txt --output_layer=final_result --image=/tf_files/satellite/11856-26822-16.jpg
I tensorflow/examples/label_image/main.cc:205] highway (0): 0.907585
I tensorflow/examples/label_image/main.cc:205] noway (1): 0.0924149

unmapped

90% chance there's an unmapped road at 16/11820/26685

Go to Wiki to see more prediction results.

Running a prediction server

% docker run -p 3000:3000 --name=inception_container -it rodowi/inception_serving

root@711a84710476:/unmapped# npm i && ./lib/server.js
Listening for tile requests in port 3000
➜  unmapped git:(serve) ✗ curl -s localhost:3000?tile=16/11820/26685 | grep "way "
2017-03-02 15:52:59.785471: I tensorflow/examples/label_image/main.cc:206] highway (0): 0.957238
2017-03-02 15:52:59.785531: I tensorflow/examples/label_image/main.cc:206] noway (1): 0.0427618

About

ConvNet finding unmapped roads in satellite imagery

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published