HCT-Det: a hybrid CNN-transformer architecture for 3D object detection from point clouds

Chao Wang

doi:10.1117/12.3005832

10 October 2023 HCT-Det: a hybrid CNN-transformer architecture for 3D object detection from point clouds

Chao Wang

Proceedings Volume 12799, Third International Conference on Advanced Algorithms and Signal Image Processing (AASIP 2023); 127993I (2023) https://doi.org/10.1117/12.3005832
Event: 3rd International Conference on Advanced Algorithms and Signal Image Processing (AASIP 2023), 2023, Kuala Lumpur, Malaysia

Abstract

Detecting 3D objects from LiDAR points is significant for the environmental perception of robotic systems. Some pillarbased 3D object detectors solely use 2D convolutions as feature encoders, which occupy fewer computation resources but sacrifice model accuracy. To activate the potential performance of pillar-based feature representation manners, we propose HCT-Det, a novel hybrid CNN-Transformer architecture for 3D object detection from point clouds. Motivated by the structure re-parameterization technique and vision transformer (ViT) framework, we redesign the 2D backbone and further introduce the Rep-VGG block and multi-head self-attention (MHSA) mechanism to enrich the scale diversity of feature representation. We perform ablation experiments on the KITTI vision benchmarks to highlight the superiority of our HCT-Det. The evaluation results show that our model outperforms PointPillars baseline, yielding an accuracy of 79.08 moderate AP_3D on the car category at a speed of 57.46 FPS on the NVIDIA Tesla P40 platform. Without bells and whistles, our HCT-Det can achieve a reasonable trade-off between accuracy and speed.

(2023) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Chao Wang "HCT-Det: a hybrid CNN-transformer architecture for 3D object detection from point clouds", Proc. SPIE 12799, Third International Conference on Advanced Algorithms and Signal Image Processing (AASIP 2023), 127993I (10 October 2023); https://doi.org/10.1117/12.3005832

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available

Members: $17.00

Non-members: $21.00 ADD TO CART

PROCEEDINGS
6 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Object detection

Point clouds

Convolution

3D modeling

LIDAR

Feature extraction

Voxels

Keywords/Phrases

Search In:

Publication Years