3D-QAE: Fully Quantum Auto-Encoding of 3D Point Clouds

Lakshika Rathi1   Edith Tretschk2   Christian Theobalt2   Rishabh Dabral2   Vladislav Golyanik2
1IIT Delhi   2MPI for Informatics, SIC
British Machine Vision Conference (BMVC) 2023
BMVC Category: "Brave new ideas"

3D-QAE, a quantum point cloud auto-encoder. We prepare a classical 3D point cloud as input and then encode it into a quantum state vector |ψin⟩ of two sets of qubits, A and B, via amplitude encoding. The encoder E (visualised here with J=1 block) acts on this state vector via a learned unitary transform implemented by a parametrised quantum circuit. At the bottleneck, we remove the information stored in the qubits B. This removal acts as a quantum non-linearity whose output is the latent vector |φ ⟩ of qubits A. We re-initialise qubits B to |0⟩ and let the decoder D, whose architecture is the same as E's, transform qubits A and B. We then measure the output of D to obtain the state vector |ξ⟩, which we can classically process in a loss function or convert to the final 3D output reconstruction.

Abstract

Existing methods for learning 3D representations are deep neural networks trained and tested on classical hardware. Quantum machine learning architectures, despite their theoretically predicted advantages in terms of speed and the representational capacity, have so far not been considered for this problem nor for tasks involving 3D data in general. This paper thus introduces the first quantum auto-encoder for 3D point clouds. Our 3D-QAE approach is fully quantum, i.e. all its data processing components are designed for quantum hardware. It is trained on collections of 3D point clouds to produce their compressed representations. Along with finding a suitable architecture, the core challenges in designing such a fully quantum model include 3D data normalisation and parameter optimisation, and we propose solutions for both these tasks. Experiments on simulated gate-based quantum hardware demonstrate that our method outperforms simple classical baselines, paving the way for a new research direction in 3D computer vision.

Downloads


  • Paper, 5.5 MB

  • Poster, 1.8 MB

  • Source Code

Citation

BibTeX, 1 KB

 
@InProceedings{Rathi2023, 
    author={Rathi, Lakshika  and Tretschk, Edith and Theobalt, Christian and Dabral, Rishabh  and Golyanik, Vladislav}, 
    title={{3D-QAE}: Fully Quantum Auto-Encoding of 3D Point Clouds}, 
    booktitle={The British Machine Vision Conference (BMVC)}, 
    year={2023} 
}

Contact

For questions, clarifications, please get in touch with
Rishabh Dabral or
Vladislav Golyanik.
This page is Zotero translator friendly. Page last updated Imprint. Data Protection.