Learning Cartographic Building Generalization with Deep Convolutional Neural Networks

Introduction

Building simplification is conducted in this work using a fully conv layers with down-conv and up-conv. The original work was provided by a master thesis, which use the car trajectories to reconstruct the road networks.

Comparison

Buildings shown at original scale 1:15,000, from top to bottom: Test 1 and Test 2; from left to right: input image, target image, U-net prediction, Residual U-net prediction and GAN prediction.

Citation:

Please cite this paper in your publications if it helps your research:

@article{feng2019learning,
  title={Learning cartographic building generalization with deep convolutional neural networks},
  author={Feng, Yu and Thiemann, Frank and Sester, Monika},
  journal={ISPRS International Journal of Geo-Information},
  volume={8},
  number={6},
  pages={258},
  year={2019},
  publisher={Multidisciplinary Digital Publishing Institute}
}

Setting for DCOS

{
  "id": "/yourname-1gpu",
  "backoffFactor": 1.15,
  "backoffSeconds": 1,
  "cmd": "",
  "container": {
    "type": "MESOS",
    "volumes": [],
    "docker": {
      "image": "tensorflow/tensorflow:1.5.0-gpu-py3",
      "forcePullImage": false,
      "parameters": []
    }
  },
  "cpus": 1,
  "disk": 0,
  "instances": 0,
  "maxLaunchDelaySeconds": 3600,
  "mem": 10000,
  "gpus": 1,
  "networks": [
    {
      "mode": "host"
    }
  ],
  "portDefinitions": [],
  "requirePorts": false,
  "upgradeStrategy": {
    "maximumOverCapacity": 1,
    "minimumHealthCapacity": 1
  },
  "killSelection": "YOUNGEST_FIRST",
  "unreachableStrategy": {
    "inactiveAfterSeconds": 0,
    "expungeAfterSeconds": 0
  },
  "healthChecks": [],
  "fetch": [],
  "constraints": []
}

Connect to DCOS instance via Windows

1. Install Docker (https://www.docker.com/get-docker ) under Windows.

2. Dowload a docker image and start an instance according to this image (z.B. tensorflow/tensorflow:1.5.0-gpu ).

• First time run it

docker pull tensorflow/tensorflow:1.5.0-gpu                   // Download a docker image from dockerhub
docker run tensorflow/tensorflow:1.5.0-gpu                    // Run this docker iamge
docker ps                                                     // See your current working instance name
winpty docker exec -it yourDockerInstanceName bash            // Run bash on this instance

• Latter run it (After you restart your PC)

docker ps  -a                                                 // See your already shut down working instance name
docker start yourDockerInstanceName                           // Start your instance of the downloaded image
winpty docker exec -it yourDockerInstanceName bash            // Run  bash on this instance

3. Install dcos tool at this local docker instance with the following code (Latter do not need to install it again)

[ -d /usr/local/bin ] || mkdir -p /usr/local/bin && 
curl https://downloads.dcos.io/binaries/cli/linux/x86-64/dcos-1.10/dcos -o dcos && 
mv dcos /usr/local/bin && 
chmod +x /usr/local/bin/dcos && 
dcos cluster setup http://130.75.51.24 && 
dcos

4. Open the GPU instance at DCOS

dcos task exec -it myname-2gpu bash

Dependencies and Settings

Install the dependcies and set up connecting folder

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/nvidia/lib64 &&
export PATH=$PATH:/usr/local/nvidia/bin:/usr/local/sbin:/usr/sbin:/sbin &&
apt-get update &&
apt-get install cifs-utils -y &&
apt-get install git -y &&
apt-get install python3-pip -y &&
pip3 install keras &&
apt-get install python3-tk -y &&
apt-get install python3-skimage -y &&
apt install gdal-bin python-gdal python3-gdal -y &&
git clone https://www.github.com/keras-team/keras-contrib.git &&
cd keras-contrib &&
python3 setup.py install &&
cd ../ &&
mkdir tmp &&
mount -t cifs -o user=,password= //130.75.51.38/tmp/yu tmp &&
cd tmp 

git clone https://www.github.com/keras-team/keras-contrib.git
cd keras-contrib
python setup.py install

Run a python script and delete the DCOS instance after it finished (Optional)

python3 simply.py
curl -X DELETE http://130.75.51.24/marathon/v2/apps/yourinstancename

Git Commands

git add -A && git commit -m "Your Message"