Public | Automated Build

Last pushed: a year ago
Short Description
Open Source Structure from Motion pipeline
Full Description



OpenSfM is a Structure from Motion library written in Python on top of OpenCV. The library serves as a processing pipeline for reconstructing camera poses and 3D scenes from multiple images. It consists of basic modules for Structure from Motion (feature detection/matching, minimal solvers) with a focus on building a robust and scalable reconstruction pipeline. It also integrates external sensor (e.g. GPS, accelerometer) measurements for geographical alignment and robustness. A JavaScript viewer is provided to preview the models and debug the pipeline.

<p align="center">
<a href="">
<img src="" />

Checkout this blog post with more demos


Installing dependencies on MacOSX

Install OpenCV using

brew tap homebrew/science
brew install opencv
brew install homebrew/science/ceres-solver
brew install boost-python
sudo pip install -r requirements.txt

And install OpenGV using

git clone
cd opengv
mkdir build
cd build
make install

Be sure to update your PYTHONPATH to include /usr/local/lib/python2.7/site-packages where OpenCV and OpenGV have been installed. For example:

export PYTHONPATH=/usr/local/lib/python2.7/site-packages:$PYTHONPATH

Installing dependencies on Ubuntu

See the Dockerfile for the commands to install all dependencies on Ubuntu 14.04. The steps are

  1. Install OpenCV, Boost Python, NumPy, SciPy using apt-get
  2. Install python requirements using pip
  3. Clone, build and install OpenGV following the receipt in the Dockerfile
  4. Build and Install the Ceres solver from its source using the -fPIC compilation flag

Install note

When running OpenSfM on top of OpenCV 3.0 the OpenCV Contrib modules are required for extracting SIFT or SURF features.


python build


An example dataset is available at data/berlin.

  1. Put some images in data/DATASET_NAME/images/
  2. Put config.yaml in data/DATASET_NAME/config.yaml
  3. Go to the root of the project and run bin/run_all data/DATASET_NAME
  4. Start an http server from the root with python -m SimpleHTTPServer
  5. Browse http://localhost:8000/viewer/reconstruction.html#file=/data/DATASET_NAME/reconstruction.meshed.json.

Things you can do from there:

  • Use datasets with more images
  • Click twice on an image to see it. Then use arrows to move between images.
  • Run bin/mesh data/berlin to build a reconstruction with sparse mesh that will produce smoother transitions from images

Thanks to sponsors

  • Thank you Jetbrains for supporting the project with free licenses for IntelliJ Ultimate. Contact peter at mapillary dot com if you are contributor and need one. Apply your own project here
Docker Pull Command
Source Repository