This project explores a novel approach to Implicit Neural Representation (INR) using a hybrid quantum-classical neural network. Traditional INRs, while effective, often struggle to capture high-frequency details in signals, leading to blurry or indistinct outputs. Our research introduces the Quantum Implicit Neural Representation (QIREN), which overcomes this limitation by leveraging the inherent power of quantum circuits.
The QIREN model uses a data re-uploading technique within its quantum circuit to exponentially enhance its ability to model the frequency spectrum of a signal. This allows the network to learn and represent fine-grained details more effectively than its classical counterparts. The project evaluates QIREN's performance on two key tasks: image regression and image super-resolution, demonstrating its superior capability in reconstructing and enhancing images.
To run this project, you need to have the following installed on your system:
- Python 3.8 or higher
- Jupyter Notebook or JupyterLab
Follow these steps to set up the project environment:
-
Clone the repository from its source.
git clone <repository_url>
-
Navigate into the project directory.
cd <project_directory>
-
Install the required Python packages.
pip install -r requirements.txt
The entire project workflow is contained within a single Jupyter Notebook. To run the experiments and see the results:
-
Start a Jupyter session from your terminal.
jupyter notebook
-
Open the notebook named
Imagefittingpaperwithsuperresolutionfinal.ipynb. -
Run all the cells in the notebook in order. The notebook will guide you through the process of setting up the model, training it for both image regression and super-resolution tasks, and displaying the comparative results with classical models.
Our experiments show that the QIREN architecture significantly outperforms a classical MLP with Random Fourier Features (RFF) encoding and ReLU activation. The superior performance of QIREN is particularly evident in the Peak Signal-to-Noise Ratio (PSNR), a key metric for image quality.
| Task | Model | PSNR (dB) |
|---|---|---|
| Image Regression | RFF + ReLU | 25.49 |
| Image Regression | Quantum Hybrid | 33.26 |
| Super-Resolution | RFF + ReLU | 20.07 |
| Super-Resolution | Quantum Hybrid | 29.83 |
The results confirm that the QIREN model's ability to capture high-frequency details translates directly into higher-quality, more detailed reconstructed images.