Cameron Braunstein, Vladislav Golyanik, Eddy Ilg
This project is the source code used for Quantum-Hybrid Stereo Matching with Nonlinear Regularization and Spatial Pyramids. This work was accepted to 3DV 2024.
This code is written in Python. We recommend conda for ease of set up, however this is not strictly necessary.
Once in our Python environment of choice, run
$ pip install -r requirements.txt
to install all necessary Python libraries
In order to have access to DWave's remote services, you must have an account with DWave, as well as some additional configuration set up on your local machine. Follow the instructions presented here.
There are additional steps needed to set up Gurobi for optimizations. Start with the guide here for set up instructions.
Optionally, one can also install iViz to view relevant data quickly. Alternatively, however, one can view the outputs via standard image viewing applications without any trouble.
The Middlebury data which we work with in our paper is all available here. Due to the small size of this data, we have included it in the repository.
If you wish to work with the Sintel stereo data, download the .zip from here, and place it into the /datasets directory. Next, unzip the file. In Linux, this can be done using the unzip command, for example
$ unzip datasets/MPI-Sintel-stereo-training-20150305.zip
To see the gurobi optimizer run on the Tsukuba pair from the Middlebury dataset, run
$ ./example_tsukuba.sh
The output (before filtering) should look like this:
For the first frame of the Market 2 scene of Sintel, run
$ ./example_sintel.sh
Here are the results of this estimation visualized with matplotlib:
One can see the result of running this estimation process across an entire Sintel scene below:
If you find this code useful for your research, please cite our paper:
@inproceedings{braunstein2023quantumhybrid,
title={Quantum-Hybrid Stereo Matching With Nonlinear Regularization and Spatial Pyramids},
booktitle={International Conference on 3D Vision (3DV)},
author={Cameron Braunstein and Eddy Ilg and Vladislav Golyanik},
year={2024}
}
This work is under the MIT License.