This repository demonstrates how to incorporate and use deep-learning based local features and matchers in COLMAP. Compared to the default SIFT features of COLMAP, it can provide denser and higher quality image matching.
convert.py
is for preparing input image data for 3D Gaussian Splatting. Place the images you wish to use in the <location>/input
directory.
<location>
|---input
|---<image 0>
|---<image 1>
|---...
To run the Custom COLMAP code, which serves as an alternative to the original implementation of convert.py, execute the following command:
python3 convert.py --source_path <location>
Run the Docker container (if necessary).
sh docker_container.sh