Project Overview
Use the chatbot at the bottom right to explore the QRF system and ask any questions!
NeRF vs QRF
- NeRF: Given a 3D position (x, y, z) and viewing direction (θ, ϕ), outputs RGB color (r, g, b) and transparency (σ).
- QRF: Replaces this with quantum encoding circuits, parameterized quantum circuits, and quantum activations.
QRF Architecture
The QRF system maps input position and viewing direction to RGB and σ using quantum components. It enforces opacity σ to be view-independent.
Optimization
QRF is trained via gradient descent using L2 loss and the Adam optimizer with learning rate α = 0.001.
Comparison
Rendering comparison of Chair, Ship, and Hotdog after 100k training iterations. QRF shows faster convergence and better quality.
Real-World Results
Camera positions reconstructed with COLMAP.
Pine Tree in Front of My Home
Training: 4k iterations
NeRF
QRF
Succulent Plan Potted
Training: 4k iterations with 16 samples.
3D mesh result with 64 samples:
Bench Park
Training: 120k iterations with 64 samples.
3D mesh result with 64 samples:
Eluanbi Lighthouse
Paper and Supplementary Material
Citation
@article{YuanFuYang2025,
title={Quantum Radiance Fields: A Fully Quantum-Native Framework for Photorealistic Neural Rendering},
author={YuanFu Yang, Min Sun},
journal={arXiv preprint arXiv:2211.03418},
year={2025}
}