Skip to content

Latest commit

 

History

History
60 lines (52 loc) · 2.11 KB

README.md

File metadata and controls

60 lines (52 loc) · 2.11 KB

UPDATE (3rd of January 2023)

New version with some improvements including much faster and more accurate essential matrix estimation (and fundamental matrix with known intrinsics) is available in the branch version2. If no issues will be raised then new branch will be merged with master.

Solvers for:

  • Affine matrix (3 points)
  • Homography matrix (4 points)
  • Fundamental matrix (7 and 8 points)
  • Essential matrix (5 points)
  • Perspective projection matrix (3 points P3P, linear 6 points)

Install and run:

  1. git submodule update --init
  2. mkdir build && cd build
  3. cmake ..
  4. make -j $(nproc)
  5. ./vsac

Install Python bindings:

python3 ./setup.py install

Run Python example:

python3 python/example.py

Libraries:

  • OpenCV (required)
  • Eigen (optional, recommended)
  • LAPACK (optional)

Note, to run essential matrix estimation either Eigen or LAPACK has to be installed.
If your own CMakeLists.txt is used then include add_definitions(-DHAVE_LAPACK) and add_definitions(-DHAVE_EIGEN) in case any of these libraries is available. Under lib directory there are Eigen headers that are included in case Eigen library is not installed, otherwise this directory can be deleted.

File with example code is samples/samples.cpp. For Python it is python/example.py. The VSAC framework enables to add new methods and solvers. To do it you need:

  1. Implement your new method as a derived class.
  2. Add a name of your method in the inlude/vsac.hpp enum.
  3. Expand if condition in the src/init.cpp to choose your method.

Older VSAC version is integrated into OpenCV could be run with the following flags:

  • USAC_PROSAC
  • USAC_ACCURATE
  • USAC_PARALLEL
  • USAC_DEFAULT
  • USAC_FAST
  • USAC_MAGSAC

Citation:

@InProceedings{Ivashechkin_2021_ICCV,
    author    = {Ivashechkin, Maksym and Barath, Daniel and Matas, Ji\v{r}{\'\i}},
    title     = {VSAC: Efficient and Accurate Estimator for H and F},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2021},
    pages     = {15243-15252}
}