The SpaceNet 7 Baseline Algorithm

The SpaceNet 7 dataset contains ~100 data cubes of monthly Planet 4 meter resolution satellite imagery taken over a two year time span, with attendant building footprint labels. The goal of the SpaceNet 7 Challenge is to identify and track building footprints and unique identifiers through the multiple seasons and conditions of the dataset.

To address this problem we propose the SpaceNet 7 Baseline algorithm. This algorithm is a multi-step process that refines a deep learning segmentation model prediction mask into building footprint polygons, and then matches building identifiers (i.e. addresses) between time steps. See The DownLinQ for further details. While the goal is akin to traditional video object tracking, the semi-static nature of building footprints and extremely small size of the objects introduces unique challenges.

There are a few steps required to run the algorithm, as detailed below.

This fork from the original repo (CosmiQ/CosmiQ_SN7_Baseline) slightly refactors the code to more closely resemble the submission format. The underlying algorithm and implementation are unchanged.

---

I. Download Data and Create Environment

1. Download SpaceNet 7 Data

   The SpaceNet data is freely available on AWS, and all you need is an AWS account and the AWS CLI installed and configured. Once you’ve done that, simply run the command below to download the training dataset to your working directory (e.g. /local_data/sn7/aws_download/):

    cd /local_data/sn7/aws_download/
    aws s3 cp s3://spacenet-dataset/spacenet/SN7_buildings/tarballs/SN7_buildings_train.tar.gz .
    aws s3 cp s3://spacenet-dataset/spacenet/SN7_buildings/tarballs/SN7_buildings_test_public.tar.gz .
   

   Extract data from the tarballs:

     cd /local_data/sn7/aws_download/
     tar -xvf SN7_buildings_train.tar.gz
     tar -xvf SN7_buildings_test_public.tar.gz
   

2. Download SpaceNet 7 baseline code to the desired location (e.g. /path_to_baseline/):

    cd /path_to_baseline/
    git clone https://github...
   

3. Build and launch the docker container, which relies upon Solaris (a GPU-enabled machine and nvidia-docker are recommended):

    nvidia-docker build -t sn7_baseline_image /path_to_baseline/docker 
    NV_GPU=0 nvidia-docker run -it -v /local_data:/local_data  -ti --ipc=host --name sn7_gpu0 sn7_baseline_image
    conda activate solaris
   

4. Execute commands

   1. Either use the jupyter notebook:

       cd /path_to_baseline/
       jupyter notebook --ip 0.0.0.0 --port=9111 --no-browser --allow-root &
      

      Locally in your browser, visit:

       http://localhost:9111
      

   2. Or simply copy the relevant commands into a terminal within the docker container

---

II. Prepare the data for training and inference (see notebooks/sn7_data_prep.ipynb)

---

III. Train (see notebooks/sn7_baseline.ipynb, or use the pretrained weights in models

---

IV. Infer (see notebooks/sn7_baseline.ipynb)

Output will consist of:

1. Inference masks:

2. Building footprints:

3. Building footprints matched by ID across the data cube: