This C++ repository implements a number of state-of-the-art algorithms for large-scale Bundle Adjustment, including
- IRLS: Iteratively Re-weighted Least Squares
- M-HQ: Multiplicative Half Quadratic Lifting.
- GOM+: Graduated Non-convexity.
- MOO: Robust fitting with Multi-Objective Optimization
- ASKER: Adaptive Kernel Scaling
- ReGeMM: Relaxed Generalized Majorization Minimization (ReGeMM)
See the following paper for more details as well as the comparisons between the methods.
The methods are written is C++, developed mainly based on the SSBA library. In this release, we use Conjugate Gradient to solver the linear system. Therefore, we can handle larger datasets compared to previous SSBA releases.
A CMake file is provided. Before compilation, please set OPTIMIZER variable (in the CMakeLists.txt file) to the desired method. For instance, use
set (OPTIMIZER schur_irls)
to compile the IRLS algorithm. The output executable will be bundle_irls
Similarly, set the OPTIMIZER variable to the following to compile the associated algorithms:
- schur_lifting: M-HQ
- schur_gnc: GOM+
- schur_gnc_moo: MOO
- schur_filter: ASKER
- schur_gemm: ReGeMM
After setting the desired method in the CMakeLists.txt file, create a build folder (from the source directory) and start compiling:
mkdir build
cd build
cmake ..
make -j8
A sample input file is provided in the Dataset folder. After compiling, run
bundle_<method> ../Dataset/ladybug-49.txt
Please create issue on this repository, or contact [email protected] , where
myemail=huul