Official repository for the paper "Quantum Multi-Model Fitting", published at CVPR 2023. Authors: Matteo Farina, Luca Magri, Willi Menapace, Elisa Ricci, Vladislav Golyanik and Federica Arrigoni.

If you use this code, or you find some of this repository helpful, please cite:

@inproceedings{Farina2023quantum, 
title={Quantum Multi-Model Fitting}, 
author={Matteo Farina and Luca Magri and Willi Menapace and Elisa Ricci and Vladislav Golyanik and Federica Arrigoni}, 
booktitle={Computer Vision and Pattern Recognition (CVPR)}, 
year={2023} 
}

This code has been tested under Ubuntu 18.04, 20.04 and 22.04 through a Miniconda virtual environment. To install all the dependencies to run the demo script demo.py with a CPU, run:

conda env create -f deps/environment.yml

You can then activate your virtual environment with conda activate qmmf.
The virtual environment comes with the DWave Ocean Software Development Kit. In order to use a DWave AQC, please follow the setup instructions at this link and this other link.

This repo contains the source code for QuMF and DeQuMF, along with their CPU counterparts QuMF (SA) and DeQuMF (SA), as described in the paper. You can find them in problems/disjoint_set_cover.py. Have a look at the main function in demo.py to instantiate each of these algorithms.

The demo script demo.py runs a qualitative and quantitative demo of DeQuMF (SA) on the AdelaideRMF dataset for fundamental matrix estimation. Qualitative results are by default stored in the demo_output folder.

IMPORTANT NOTE: Commented code for QuMF is given in order to show how to use this algorithm, too. However, the user should bear in mind the considerations reported in Section 5 of the paper: given the current state of Adiabatic Quantum Computers, QuMF cannot support large problems as the ones resulting from sampling the AdelaideRMF dataset. Your demo.py run will likely crash if you select QuMF as it is impossible to perform minor embedding.

The datasets we used for our paper are linked here:

1. AdelaideRMF for Fundamental Matrix estimation: link;
2. Traffic2 and Traffic3 subsets of the Hopkins benchmark: link;
3. York Urban Line Segment Databse: link.

Details on the parameter settings can be found in the Supplementary Material of our paper, containing information on the number of sampled models and the used inlier thresholds, too.

This paper is supported by FAIR (Future Artificial Intelligence Research) project, funded by the NextGenerationEU program within the PNRR-PE-AI scheme (M4C2, Investment 1.3, Line on Artificial Intelligence). This work was partially supported by the PRIN project LEGO-AI (Prot. 2020TA3K9N) and it was carried out in the Vision and Learning joint laboratory of FBK and UNITN.