⚠️ This repository has been archived, mainly because it relied on GitHub Actions using a deprecated Ubuntu 18.04 image to build and distribute a Docker image with preinstalled software. Additionally, this project was mostly educational and not intended for production use.

I’ve learned a lot about photogrammetry and continuous integration since I started writing this pipeline in 2020, during the COVID pandemic. However, today we have better and more modern tools for this kind of task (see my GitHub profile, for example).

If you’re still interested in this particular pipeline, you can check out .github/workflows/release.yml as a reference. Also, there’s a chance I might resurrect this pipeline on GitLab if I find the time.

linux-photogrammetry-tools

A set of open-source photogrammetry tools to run with Docker:

• SIFT* by vlfeat.org.
• Bundler (compiled w/Ceres Solver) by Noah Snavely.
• CMVS & PMVS2 by Yasutaka Furukawa.
• A patched version of Isaac Lenton's bundler.py:
  • Ported to Python 3.
  • Works with VLFeat SIFT (patch based on Python Photogrammetry Toolbox code).
  • Uses Ceres Solver by default.
  • Reads CCD_WIDTHS from a YAML file.
• An image resizer script that keeps EXIF metadata.
• A Makefile to run the pipeline steps.
• A Jupyter Notebook for point cloud meshing with Open3D.

If you want to know more about how this software is packaged see here.

* SIFT patent expired on March 2020.

Usage

Options

The following options can be passed to make in the command line:

• IMG_DIR: path. Directory with a collection of images to be processed. Default: examples/kermit.
• RESIZE: bool. Resize pictures before processing. Default: True.
• MAX_SIZE: int. If RESIZE is True, maximum size in pixels (width or height). Default: 1200.
• LOGFILE: path. Default: log.txt.

Visualize results & post-processing

Open3D

To generate a 3D mesh from the point cloud with Open3D you will need Anaconda or Miniconda installed on your system.

$ conda create -n open3d -c open3d-admin -c conda-forge open3d=0.12

Finally, run the notebook gen_3d_mesh.ipynb located at the docs folder.

Meshlab

Open the .ply file located at work_dir/pmvs/models/ with Meshlab. You should visualize the point cloud clearly. To generate a textured mesh from point cloud, I recommend following the Shubham Wagh's guide.

Known issues

1. No CCD width available for camera.

   Only a small number of CCD widths are listed in cfg/ccd_widths.yml.

   Solution: google your camera specs and add a new entry to the list.

If your problem is not listed here please check the log file inside your work_dir directory and open a new issue.

License

Code released under the GNU GPLv3 License.