Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fast Quantum Convolutional Neural Networks for Low-Complexity Object Detection in Autonomous Driving Applications

Hankyul Baek, Donghyeon Kim, Joongheon Kim

TL;DR

The paper tackles the data- and compute-driven bottlenecks of object detection in autonomous driving by proposing fast quantum convolution and a quantum object-detection workflow (QCOD). It introduces channel uploading and channel reconstruction to pack multiple input channels into a small number of qubits, plus a quantum region proposal network trained with heterogeneous knowledge distillation from a classical RPN; complexity analyses highlight potential quantum speedups with QRAM. Empirical results on KITTI show that QCOD can achieve meaningful accuracy with reasonable quantum resources (e.g., MAP ≈ 51.2 with 64 channels) and that channel uploading and KD contribute to performance gains, while highlighting challenges like information overlap at high channel counts. Collectively, the work demonstrates a feasible pathway toward quantum-accelerated object detection and motivates further hardware and algorithmic development in quantum machine learning for autonomous driving.

Abstract

Spurred by consistent advances and innovation in deep learning, object detection applications have become prevalent, particularly in autonomous driving that leverages various visual data. As convolutional neural networks (CNNs) are being optimized, the performances and computation speeds of object detection in autonomous driving have been significantly improved. However, due to the exponentially rapid growth in the complexity and scale of data used in object detection, there are limitations in terms of computation speeds while conducting object detection solely with classical computing. Motivated by this, quantum convolution-based object detection (QCOD) is proposed to adopt quantum computing to perform object detection at high speed. The QCOD utilizes our proposed fast quantum convolution that uploads input channel information and re-constructs output channels for achieving reduced computational complexity and thus improving performances. Lastly, the extensive experiments with KITTI autonomous driving object detection dataset verify that the proposed fast quantum convolution and QCOD are successfully operated in real object detection applications.

Fast Quantum Convolutional Neural Networks for Low-Complexity Object Detection in Autonomous Driving Applications

TL;DR

The paper tackles the data- and compute-driven bottlenecks of object detection in autonomous driving by proposing fast quantum convolution and a quantum object-detection workflow (QCOD). It introduces channel uploading and channel reconstruction to pack multiple input channels into a small number of qubits, plus a quantum region proposal network trained with heterogeneous knowledge distillation from a classical RPN; complexity analyses highlight potential quantum speedups with QRAM. Empirical results on KITTI show that QCOD can achieve meaningful accuracy with reasonable quantum resources (e.g., MAP ≈ 51.2 with 64 channels) and that channel uploading and KD contribute to performance gains, while highlighting challenges like information overlap at high channel counts. Collectively, the work demonstrates a feasible pathway toward quantum-accelerated object detection and motivates further hardware and algorithmic development in quantum machine learning for autonomous driving.

Abstract

Spurred by consistent advances and innovation in deep learning, object detection applications have become prevalent, particularly in autonomous driving that leverages various visual data. As convolutional neural networks (CNNs) are being optimized, the performances and computation speeds of object detection in autonomous driving have been significantly improved. However, due to the exponentially rapid growth in the complexity and scale of data used in object detection, there are limitations in terms of computation speeds while conducting object detection solely with classical computing. Motivated by this, quantum convolution-based object detection (QCOD) is proposed to adopt quantum computing to perform object detection at high speed. The QCOD utilizes our proposed fast quantum convolution that uploads input channel information and re-constructs output channels for achieving reduced computational complexity and thus improving performances. Lastly, the extensive experiments with KITTI autonomous driving object detection dataset verify that the proposed fast quantum convolution and QCOD are successfully operated in real object detection applications.
Paper Structure (17 sections, 3 theorems, 12 equations, 9 figures, 2 tables)

This paper contains 17 sections, 3 theorems, 12 equations, 9 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Quantum Machine Learning
  4. Fast Quantum Convolutional Neural Networks
  5. Motivation
  6. Architecture of fast quantum convolutional neural network
  7. Strategy of fast quantum convolution
  8. Quantum Object Detection
  9. Motivation of quantum region proposal network
  10. Architecture of quantum region proposal network
  11. Heterogeneous knowledge distillation training
  12. Loss of QRPN
  13. Implementation details
  14. Performance Evaluation
  15. Setup
  16. ...and 2 more sections

Key Result

Lemma 1

(Advantages of QRAM) Let target input $P \in \mathbb{R}^{n \times d}$, there exists a QRAM structure that conducts inserting, deleting, and updating each datum $p_{i,j}$ in time $\mathcal{O}(\log(n^2))$. In addition, there exists a quantum algorithm $|i\rangle_{address}|0\rangle_{data} \rightarrow |

Figures (9)

  • Figure 1: A brief illustration of quantum computing and its application.
  • Figure 2: Comparison between existing channel uploading strategies (a-c) and our proposed channel uploading strategy (d).
  • Figure 3: An illustration of the proposed fast quantum convolutional neural network.
  • Figure 4: Detailed illustration of the proposed fast quantum convolution.
  • Figure 5: Proposed QRPN with our fast quantum convolution (The green and orange quantum convolution filter is designed for classification and box regression. The purple quantum convolution filter is designed for 2-dimensional convolution).
  • ...and 4 more figures

Theorems & Definitions (4)

  • Lemma 1
  • Lemma 2
  • Theorem 1
  • proof