OpenMVG
github.com
- Free plan
- Open source
- Windows
- macOS
- Linux
- iPhone and iPad
- Android
- API
- Self-hosted
What OpenMVG is
OpenMVG is a multiple view geometry library for 3D computer vision and structure-from-motion work. It provides an end-to-end process for reconstructing 3D scenes from images, using libraries, binaries, and pipelines. Its libraries cover image manipulation, feature description and matching, feature tracking, camera models, multiple-view geometry, robust estimation, and structure-from-motion algorithms. Binaries can initialize scenes, detect and match features, build structure-from-motion reconstructions, and export scenes for other multiple-view-stereo frameworks. Those frameworks can compute dense point clouds or textured meshes from the exported scenes. Written in C++, OpenMVG is licensed under the Mozilla Public License 2.0. The project describes its research aim as supporting reproducible work through readable implementations of classic and state-of-the-art algorithms. It is designed to be read, learned from, modified, and used, and includes test-driven development and samples. Tutorials cover data structures, library use, and processing an image dataset. OpenMVG is free and open source, with a desktop deployment listing and platforms including Windows, macOS, Linux, iOS, and Android. The project provides a contact email and points users to its wiki for tutorials and usage resources.
Who it suits
OpenMVG suits developers and researchers working on 3D reconstruction from images or structure-from-motion workflows. Its tutorials and stated focus on readable, modifiable implementations may also suit people learning these methods.
Good to know, and what to watch
- End-to-end image-based 3D reconstruction framework.
- Libraries cover feature matching and camera models.
- Exports scenes for dense point clouds or textured meshes.
- Includes tutorials, samples, and test-driven development.
- Free under the Mozilla Public License 2.0.
- The project is developed in C++.
- Dense point clouds and textured meshes use other frameworks.
- Listed deployment is desktop.
Plans
Every plan and priceOpenMVG
Open source
The verdict
OpenMVG provides a broad set of libraries and pipeline tools for image-based 3D reconstruction. It is a free, open-source option for developers and researchers who can work with a C++ project and related multiple-view-stereo frameworks.
What the maker publishes
- Build prerequisites
- Official build instructions require CMake, Git and a C or C++ compiler; optional Qt support is version 5.4 or newer. source
- Camera localization
- The project provides image-localization functionality for estimating poses against an existing reconstruction. source
- Community footprint
- The GitHub repository page showed about 6.6 thousand stars and 1.7 thousand forks when accessed. source
- Compiler baseline
- The build guide calls for C++11-capable toolchains including Visual Studio 2015 or newer, GCC 4.8.1 or newer, or Clang 3.3 or newer. source
- Conan packages
- ConanCenter provides openmvg 2.0 packages for Linux, Windows, macOS, macOS Apple Silicon and Windows ARM64. source
- Core capabilities
- Its libraries include image manipulation, feature description and matching, feature tracking, camera models, multiple-view geometry, robust estimation, and structure-from-motion algorithms. source
- Core libraries
- The library includes image, numeric, feature, camera, multiview, linear-programming, robust-estimation, geometry, geodesy and SfM modules. source
- Design and testing
- The project describes OpenMVG as designed to be easy to read, learn, modify, and use, with test-driven development and samples. source
- Documentation
- The project links to tutorials covering its data structures, library use, and processing an image dataset. source
- Documentation depth
- The Read the Docs material documents individual binaries, data structures, camera models, solvers, pipelines and export tools, while marking some modules as incomplete. source
- GPU feature library
- POPART lists PopSIFT, a GPU SIFT implementation reported at 25 frames per second on HD images with recent graphics cards. source
- Ground-control workflows
- The release history records support for ground-control-point registration and geodesy-related workflows. source
- Implementation
- OpenMVG is developed in C++. source
- Independent praise: capability
- An independent photogrammetry tester described the resulting OpenMVG/OpenMVS reconstruction as good and substantially complete. source
- Independent praise: quality
- The same test produced a textured model the author called very pleasing, with OpenMVG taking 6 minutes 43 seconds for its own processing on that test. source
- Latest release
- Release v2.1 Sablefish was published on 28 December and recorded 42 pull requests and 45 issue or feature enhancements. source
- Licence
- The repository distributes the software under the Mozilla Public License 2.0. source
- License
- The repository identifies the project license as Mozilla Public License 2.0. source
- Limit: algorithm complexity
- A NAVER Labs tutorial said OpenMVG combines many pre-existing algorithms in ways that are impressive but make the resulting algorithm difficult to understand. source
- Limit: command-line workflow
- The same user noted that OpenMVG consists of many command-line tools and is easier to operate through custom scripts. source
- Limit: dense reconstruction dependency
- The tester reported that dense reconstruction was the difficult stage and encountered OpenMVS crashes when GPU memory demand became too high. source
- Limit: installation effort
- An independent user reported that openMVG had no prepackaged installation at the time and required manual compilation. source
- Limit: speed
- A comparative test found OpenMVG feature extraction and reconstruction slower than COLMAP for that dataset, although final model quality was similar. source
- Marker integration
- The related CCTag library detects concentric-circle markers designed for sub-pixel accuracy under blur, occlusion and poor lighting. source
Checked against the maker's pages on 15 Oct 2026.