pymagsac

This repository contains an Python wrapper of MAGSAC. Daniel Barath, Jana Noskova and Jiri Matas; MAGSAC: Marginalizing sample consensus, CVPR 2019

https://github.com/danini/magsac -- main repo.

If you use the algorithm, please cite

@inproceedings{barath2019magsac,
	author = {Barath, Daniel and Matas, Jiri and Noskova, Jana},
	title = {MAGSAC: marginalizing sample consensus},
	booktitle = {Conference on Computer Vision and Pattern Recognition},
	year = {2019},
}

@inproceedings{barath2019magsacplusplus,
	author = {Barath, Daniel and Noskova, Jana and Ivashechkin, Maksym and Matas, Jiri},
	title = {MAGSAC++, a fast, reliable and accurate robust estimator},
	booktitle = {arXiv preprint:1912.05909},
	year = {2019},
}

If you use it for fundamental matrix estimation with DEGENSAC turned on, please cite

@inproceedings{Chum2005,
  author = {Chum, Ondrej and Werner, Tomas and Matas, Jiri},
  title = {Two-View Geometry Estimation Unaffected by a Dominant Plane},
  booktitle = {CVPR},
  year = {2005},
}

Performance

Magsac is the state of the art according to the recent study Yin et.al."Image Matching across Wide Baselines: From Paper to Practice", 2020.

Requirements

• Python 3
• CMake 2.8.12 or higher
• Eigen3
• libgcc-ng
• OpenCV 3.4
• A modern compiler with C++17 support

Preparation for conda env.

Activate conda env

conda activate torch

Install cmake

conda install -c conda-forge cmake

Install eigen3

conda install -c omnia eigen3

Install libgcc

conda install -c conda-forge libgcc-ng

Installation

To build and install python_cpp_example, clone or download this repository, copy estimators.h from the magsac repo, update cmake file, and run setpu.py install as follows:

git clone https://github.com/cognitiveRobot/magsac.git
git clone https://github.com/cognitiverobot/pymagsac
cp magsac/include/estimators.h pymagsac/src/pymagsac/include/
sed -i 's@set(CMAKE_CXX_STANDARD 14)@set(CMAKE_CXX_STANDARD 17)@' pymagsac/CMakeLists.txt
cd pymagsac
python setup.py install

or

pip3 install .

Example of usage

import pymagsac
H, mask = pymagsac.findHomography(src_pts, dst_pts, 3.0)

#  findFundamentalMatrix arguments:
	# (x1y1: numpy.ndarray[float64], 
	# x2y2: numpy.ndarray[float64], 
	# use_magsac_plus_plus: bool=True, 
	# sigma_th: float=1.0, 
	# conf: float=0.99, 
	# max_iters: int=10000, 
	# partition_num: int=5)
F, mask = pymagsac.findFundamentalMatrix(src_pts, dst_pts, 3.0)

See also this notebook

Acknowledgements

This wrapper part is based on great Benjamin Jack python_cpp_example.