colmap_cuda_ffmpeg

Docker container for running COLMAP and GLOMAP using CUDA and ffmpeg

Batch reconstruction from a bunch of movies

You can use the process_movies.bsh to loop through all the mov files in a directory and automatically run the reconstruction process on all of them.

Running it

The host machine needs to have nvidia drivers installed.

• Clone the repo
• Put your movie in the "video" directory
• Set the correct paths and set the desired FPS for frame extraction in config/colmap_config.ini
• Make sure the "images" directory is empty
• Run it as follows

docker run --gpus all  \
  -v "$(pwd)/video":/workspace/video  \
  -v "$(pwd)/images":/workspace/images  \
  -v "$(pwd)/output":/workspace/output   \
  -v "$(pwd)/config":/workspace/config   \
  -e CONFIG_PATH=/workspace/config/colmap_config.ini \
  thomashollier/colmap_ffmpeg

If you do not provide a "video" volume, the container will use any image already existing in "images" so you an curate the input images rather than extracting video frames

docker run --gpus all  \
  -v "$(pwd)/images":/workspace/images  \
  -v "$(pwd)/output":/workspace/output   \
  -v "$(pwd)/config":/workspace/config   \
  -e CONFIG_PATH=/workspace/config/colmap_config.ini \
  thomashollier/colmap_ffmpeg

Building it

docker build -t [USERNAME]/colmap_ffmpeg .