Q-Seg is a novel unsupervised image segmentation method leveraging quantum annealing, specifically designed for the existing D-Wave Advantage quantum hardware. It addresses pixel-wise segmentation by formulating it as a graph-cut optimization task. Q-Seg stands out for its scalability and efficiency, surpassing classical methods in runtime performance and offering competitive segmentation quality, particularly in Earth Observation image analysis.
Figure: Overview of the Q-Seg algorithm.
- Unsupervised segmentation using quantum annealing.
- Efficient use of D-Wave's qubit topology for scalability.
- Superior runtime performance compared to classical optimizers like Gurobi.
- Effective in scenarios with limited labeled data, such as flood mapping and forest coverage detection.
The process of Q-Seg utilizing the D-Wave quantum annealer is illustrated in the figure below. The pipeline consists of several key steps:
- Converting the Image to Graph: The image is first translated into an undirected, weighted, connected grid-graph wtih the pixels correspond to vertices and the edge weights are assigned between neighboring pixels using an appropriate similarity metric.
- Reformulating the Minimum Cut Problem: The segmentation task is converted into a Quadratic Unconstrained Binary Optimization (QUBO) problem.
- Remote Access Authentication: A private token is used for secure remote access to the D-Wave quantum annealer.
- Qubit Mapping: The
minorminer
tool maps the logical qubits in the QUBO to the physical qubits in the D-Wave hardware. - Problem Submission and Queuing: The problem instance is transmitted over the internet and queued on the shared D-Wave device.
- Quantum Annealing: The Quantum Processing Unit (QPU) on the D-Wave device performs the annealing process.
- Solution Extraction: The retrieval of the sample set produced by the QPU and extracting the optimal solution based on the lowest energy state.
- Solution Decode: The final step involves lowest energy binary solution string being decoded into a binary segmentation mask.
Figure: Operational pipeline of Q-Seg using D-Wave quantum annealer.
- Python 3.9.11
- Jupyter Notebook
- Required Python packages:
numpy
,networkx
,qiskit
,dwave-system
Clone the repository and install the required packages:
git clone https://github.com/yourusername/Q-seg.git
cd Q-seg
pip install -r requirements.txt
Navigate to the examples
directory and open the Jupyter Notebook:
cd examples
jupyter notebook q_seg_pipeline.ipynb
Follow the instructions in the notebook to perform image segmentation on a sample grayscale image.
Q-Seg/
│
├── qseg/
│ ├── __init__.py
│ ├── graph_utils.py
│ ├── dwave_utils.py
│ └── utils.py
│
├── examples/
│ ├── q_seg_pipeline.ipynb
│ ├── Annealer_vs_Gurobi_experiments.ipynb
│ └── plots.ipynb
│
├── data/
│ ├── sample_image.jpg
│ └── results/ # Logged results from experiments
│
├── figures/
│ └── operational_pipeline.png
│ └── overview_pipeline.png
│
├── requirements.txt
└── README.md
- graph_utils.py: Contains
image_to_grid_graph
anddraw
functions for graph-related operations. - dwave_utils.py: Includes
dwave_solver
andannealer_solver
functions, dedicated to D-Wave specific operations and quantum annealing. - utils.py: Features the
decode_binary_string
function, useful for general utility purposes across the project.
- q_seg_pipeline.ipynb: Located in the
examples
directory, this Jupyter notebook demonstrates the usage of Q-Seg functions on a toy example, providing a step-by-step walkthrough of the algorithm.
-Annealer_vs_Gurobi_experiments.ipynb: Contains code to run experiments on a synthetic dataset.
- plots.ipynb: Used for parsing logged data and generating plots as shown in the paper.
- Store specific data and results in the
data
directory, including a subdirectoryresults
with our experiment results for different seeds. - Place figures and diagrams in the
figures
directory.
- List all required dependencies in the
requirements.txt
file, such asqiskit_optimization
,pylatexenc
,dwave-ocean-sdk
, andgurobipy
.
If you find Q-Seg useful in your research, please consider citing our paper:
@article{venkatesh2023q,
title={Q-Seg: Quantum Annealing-based Unsupervised Image Segmentation},
author={Venkatesh, Supreeth Mysore and Macaluso, Antonio and Nuske, Marlon and Klusch, Matthias and Dengel, Andreas},
journal={arXiv preprint arXiv:2311.12912},
year={2023}
}
This project is licensed under the GNU General Public License v2.1 - see the LICENSE file for details.
For any inquiries, please reach out to:
- Supreeth Mysore Venkatesh: [email protected]
- Antonio Macaluso: [email protected]