Skip to content
No description, website, or topics provided.
HTML Jupyter Notebook Python Other
Branch: master
Clone or download
Pull request Compare This branch is 9 commits ahead of avanetten:master.
avanetten Merge pull request #8 from avanetten/master
Updates to readme and inference scripts
Latest commit a3751b5 Oct 17, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
cresi Updates to inference scripts Oct 17, 2019
docker update to v0.3 Oct 3, 2019
ipynb clean up Oct 3, 2019
results update readme Oct 3, 2019
.gitignore Create .gitignore Oct 3, 2019
LICENSE.txt update to v0.3 Oct 3, 2019
MANIFEST.in update to v0.3 Oct 3, 2019
README.md Update README.md Oct 17, 2019

README.md

Alt text

CRESI

City-scale Road Extraction from Satellite Imagery

This repository provides an end-to-end pipeline to train models to detect road networks over entire cities, and also provide speed limits and travel time estimates for each roadway. We have observed success with both SpaceNet imagery and labels, as well as Google satellite imagery with OSM labels. The repository consists of pre-processing modules, deep learning segmentation model (based upon the winning SpaceNet 3 submission by albu), post-proccessing modules to extract the road networks, inferred speed limits, and travel times. Furthermore, we include modules to scale up network detection to the city-scale, rather than just looking at small image chips. For further details see:

  1. Our arXiv paper
  2. Blogs:
    1. Large Road Networks
    2. Road Speeds
    3. OSM+Google Imagery
    4. SpaceNet 5 Baseline Part 1 - Data Prep
    5. SpaceNet 5 Baseline Part 2 - Segmentation
    6. SpaceNet 5 Baseline Part 3 - Road Graph + Speed

Install

  1. Download this repository

  2. Build docker image

     nvidia-docker build -t cresi /path_to_cresi/docker
    
  3. Create docker container (all commands should be run in this container)

     nvidia-docker run -it --rm -ti --ipc=host --name cresi_image cresi
    

Prep

  1. Prepare train/test data, e.g.:

     python /path_to_cresi/cresi/data_prep/speed_masks.py
    
  2. Edit .json file to select desired variables and point to appropriate directories


Train

  1. All at once

     cd /path_to_cresi/cresi
    
     ./train.sh jsons/sn5_baseline.json
    
  2. Run commands individually

    A. Generate folds (within docker image)

     python /path_to_cresi/cresi/00_gen_folds.py jsons/sn5_baseline.json
    

    B. Run train script (within docker image)

     python /path_to_cresi/cresi/01_train.py jsons/sn5_baseline.json --fold=0
    

Test

  1. All at once

     cd /path_to_cresi/cresi
    
     ./test.sh jsons/sn5_baseline.json
    
  2. Run commands individually

    A. Execute inference (within docker image)

     python /path_to_cresi/cresi/02_eval.py jsons/sn5_baseline.json
    

    B. Merge predictions (if required)

     python /path_to_cresi/cresi/03a_merge_preds.py jsons/sn5_baseline.json
    

    C. Stitch together mask windows (if required)

     python /path_to_cresi/cresi/03b_stitch.py jsons/sn5_baseline.json
    

    D. Extract mask skeletons

     python /path_to_cresi/cresi/04_skeletonize.py jsons/sn5_baseline.json
    

    E. Create graph

     python /path_to_cresi/cresi/05_wkt_to_G.py jsons/sn5_baseline.json
    

    F. Infer road travel time and speed limit

     python /path_to_cresi/cresi/06_infer_speed.py jsons/sn5_baseline.json
    
You can’t perform that action at this time.