Simulator HC
Regression-based Online Simulation of Starting Problem-Solution Pairs for Homotopy Continuation in Geometric Vision

CVPR 2025 (Highlight) | Paper

Overview: We propose simulatorHC, a novel regressor-simulator paradigm for geometric vision problems. It enables homotopy continuation to efficiently track only one solution path, leading to high efficiency and high success rate.

Citation

@article{zhang2024homotopy,
  title={Homotopy Continuation Made Easy: Regression-based Online Simulation of Starting Problem-Solution Pairs},
  author={Zhang, Xinyue and Dai, Zijia and Xu, Wanting and Kneip, Laurent},
  journal={arXiv preprint arXiv:2411.03745},
  year={2024}
}

Installation

To start with, install the python environment and activate it.

conda env create -f environment.yml
conda activate HC

Then install simulatorHC, remember to have cpp compiler, cmake and intel mkl toolkit installed.

pip install -e .

Run simulatorHC on Generalized Absolute Pose

# train the regressor
python simulatorHC/GeneralizedAbsolutePose/train.py
# run simulator HC with trained regressor
python simulatorHC/GeneralizedAbsolutePose/evaluate.py

Run simulatorHC on Generalized Relative Pose and Scale

# train the regressor
python simulatorHC/GeneralizedRelativePoseAndScale/train.py
# run simulator HC with trained regressor
python simulatorHC/GeneralizedRelativePoseAndScale/evaluate.py