This github repo aims to help you build a docker image with colmap residing in.

Prerequites: make sure docker and nvidia-docker2 are installed on your system. And also make sure that the nvidia driver installed on your system is compatible with cuda-8.0; otherwise you need to modify the base image in the Dockerfile.

The workflow is as below:

• switch to the directory 'colmap_in_docker' after you git-clone the repo.
• run './build_colmap.sh'; this build a docker image called 'kai46/colmap:latest' on your local machine.
• run './run_colmap.sh'; this will launch a container and lead you to a interactive bash; the script takes care of mapping host user id, and mounting host volume into the newly launched container; this script is tailored to the phoenix cluster at Cornell, and by default, it mounts the '/phoenix' directory into the launched container; feel free to modify it for your own purposes.
• enjoy!

Feel free to modify the Dockerfile and the two handy scripts yourlf. Inside the 'reference' folder, there are two Dockerfiles provided by NVIDIA. I was referencing them to figure out how to install CUDA.

There's also a pre-built docker image hosted on DcokerHub. Run 'docker pull kai46/colmap:latest' to pull the image to your local machine.