OpenSfM

Structure from Motion library written in Python on top of OpenCV.

Dependencies

• OpenCV
• libmv (included)
• ceres
• jsoncpp
• jhead
• NumPy, networkx, PyYAML

On MacOSX, use

brew tap homebrew/science
brew info opencv

It will show information about opencv package. To install, usually brew install opencv is sufficient, but to enable import from python modules, you should read 'Caveats' section saying:

    ==> Caveats
    Python modules have been installed and Homebrew's site-packages is not
    in your Python sys.path, so you will not be able to import the modules
    this formula installed. If you plan to develop with these modules,
    please run:
      mkdir -p /Users/someuser/Library/Python/2.7/lib/python/site-packages
      echo 'import site; site.addsitedir("/usr/local/lib/python2.7/site-packages")' >> /Users/someuser/Library/Python/2.7/lib/python/site-packages/homebrew.pth

Run these commands and OpenCV will be added to python site packages load path.

Then:

brew install homebrew/science/ceres-solver
brew tap cuber/homebrew-jsoncpp
brew install jsoncpp
brew install jhead
sudo pip install -r requirements.txt

Building

1. cd lib; mkdir build; cd build
2. cmake ../src
3. make

Running

An example dataset is available at data/berlin.

1. put some images in data/DATASET_NAME/images/
2. put config.yaml at data/DATASET_NAME/config.yaml
3. run bin/run_all DATASET_NAME
4. start a http server with python -m SimpleHTTPServer
5. browse http://localhost:8000/viewer/reconstruction.html#DATASET_NAME/reconstruction.json