COLMAP
Structure-from-Motion and Multi-View Stereo
COLMAP is a general-purpose, end-to-end image-based 3D reconstruction pipeline(i.e., Structure-from-Motion (SfM) and Multi-View Stereo (MVS)) with a graphicaland command-line interface. It offers a wide range of features forreconstruction of ordered and unordered image collections. The software runsunder Windows, Linux and Mac on regular desktop computers or computeservers/clusters. COLMAP is licensed under the BSD License.
If you use this project for your research, please cite the papers:
[Paper|Bibtex]and[Paper|Bibtex]
Documentation and Support
The source code is available onGitHub and documentation is available atcolmap.github.io. Please, useGitHub Discussions for questions andGitHub Issues for bug reports, feature requests/additions, etc.
Installation
Notice: For dense reconstruction and multi-GPU feature extraction/matching, compile or download the binaries with CUDA support. This requires a CUDA-enabled NVidia GPU.
→ Download Pre-Release Binaries
This is the development with all the latest improvements. In general, it is recommended to use the latest pre-release version, since it is more frequently updated with new fixes than the release version.
→ Download Release Binaries
This is the stable version, which is typically released after major new features have been added and tested for longer periods of time.
→ Build from Source
Building the latest version of COLMAP from source is the recommended approach.
→ Changelog
Vocabulary Trees
Vocabulary trees are used for scalable matching of larger datasets. The following trees are pre-built on the Flickr100k dataset. Alternatively, you can use COLMAP to build a custom vocabulary tree tuned for your target application.
- Vocabulary tree with 32K visual words
(for small-scale image collections, i.e. 100s to 1,000s of images) - Vocabulary tree with 256K visual words
(for medium-scale image collections, i.e. 1,000s to 10,000s of images) - Vocabulary tree with 1M visual words
(for large-scale image collections, i.e. 10,000s - 100,000s of images)
Datasets
A number of different test datasets are available for downloadhere with more information in thedocumentation.