Home

Awesome

This repository is no longer being updated. Future development of code tools for geospatial machine learning analysis will be done at https://github.com/cosmiq/solaris.

SpaceNet Utilities

This repository has two python packages, geoTools and evalTools. The geoTools packages is intended to assist in the preprocessing of SpaceNet satellite imagery data corpus hosted on SpaceNet on AWS to a format that is consumable by machine learning algorithms. The evalTools package is used to evaluate the effectiveness of object detection algorithms using ground truth. This is version 2.0 and has been updated with more capabilities to

Download Instructions

Further download instructions for the SpaceNet Dataset can be found here

Evaluation Metric

The evaluation metric for this competition is an F1 score with the matching algorithm inspired by Algorithm 2 in the ILSVRC paper applied to the detection of building footprints. For each building there is a geospatially defined polygon label to represent the footprint of the building. A SpaceNet entry will generate polygons to represent proposed building footprints. Each proposed building footprint is either a “true positive” or a “false positive”.

There is at most one “true positive” per labeled polygon. The measure of proximity between labeled polygons and proposed polygons is the Jaccard similarity or the “Intersection over Union (IoU)”, defined as:

alt text

The value of IoU is between 0 and 1, where closer polygons have higher IoU values.

The F1 score is the harmonic mean of precision and recall, combining the accuracy in the precision measure and the completeness in the recall measure. For this competition, the number of true positives and false positives are aggregated over all of the test imagery and the F1 score is computed from the aggregated counts.

For example, suppose there are N polygon labels for building footprints that are considered ground truth and suppose there are M proposed polygons by an entry in the SpaceNet competition. Let tp denote the number of true positives of the M proposed polygons. The F1 score is calculated as follows:

alt text

The F1 score is between 0 and 1, where larger numbers are better scores.

Hints:

python python/evaluateScene.py /path/to/SpaceNetTruthFile.csv \
                               /path/to/SpaceNetProposalFile.csv \
                               --resultsOutputFile /path/to/SpaceNetResults.csv

Data Transformation Code

To make the Spacenet dataset easier to use we have created a tool createDataSpaceNet.py This tool currently supports the creation of datasets with annotation to support 3 Formats

  1. PASCAL VOC2012
  2. Darknet
  3. Segmenation Boundaries Dataset (SBD)

It will create the appropriate annotation files and a summary trainval.txt and test.txt in the outputDirectory

Create an PASCAL VOC2012 Compatiable Dataset

The final product will have image dimensions of 420 pixels

python python/createDataSpaceNet.py /path/to/spacenet_sample/AOI_2_Vegas_Train/ \
           --srcImageryDirectory RGB-PanSharpen
           --outputDirectory /path/to/spacenet_sample/annotations/ \
           --annotationType PASCALVOC2012 \
           --imgSizePix 400

Changing the raster format

Some GIS Images have 16-bit pixel values which openCV has trouble with. createDataSpaceNet.py can convert the 16bit GeoTiff to an 8bit GeoTiff or 8bit JPEG

To create the 8bit GeoTiff

python python/createDataSpaceNet.py /path/to/spacenet_sample/AOI_2_Vegas_Train/ \
           --srcImageryDirectory RGB-PanSharpen
           --outputDirectory /path/to/spacenet_sample/annotations/ \
           --annotationType PASCALVOC2012 \
           --convertTo8Bit \
           --outputFileType GTiff \
           --imgSizePix 400
    

To create the 8bit JPEG

python python/createDataSpaceNet.py /path/to/spacenet_sample/AOI_2_Vegas_Train/ \
           --srcImageryDirectory RGB-PanSharpen
           --outputDirectory /path/to/spacenet_sample/annotations/ \
           --annotationType PASCALVOC2012 \
           --convertTo8Bit \
           --outputFileType JPEG \
           --imgSizePix 400

For more Features

python python/createDataSpaceNet.py -h

Use our Docker Container

We have created two Docker files at /docker/standalone/cpu and /docker/standalone/gpu These Dockerfiles will build a docker container with all packages neccessary to run the package

More documenation to follow

Dependencies

All dependencies can be found in the docker file Dockerfile

License

See LICENSE.