/spacenet-detection

Primary LanguageJupyter Notebook

spacenet-detection

Problem

In this project we intend to build a neural network for segmentation of building footprints in high-resolution satellite imagery using SpaceNet dataset.

Dataset

Using the Spacenet Dataset for Paris available on AWS. Download this using awscli

aws s3api get-object --bucket spacenet-dataset \
    --key AOI_3_Paris/processedData/processedBuildingLabels.tar.gz \
    --request-payer requester processedBuildingLabels.tar.gz

Annotations can be generated by cloning https://github.com/SpaceNetChallenge/utilities and running

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

Final model code is contained in stanfordmodel.ipynb

Evaluation Metric

The ground truth for objects is given as polygons. Thus we intend to use intersection over Union as the evaluation metric for object segmentation.

Main Code file

stanfordmodel.ipynb