Article: DCCA: 3D object detection with dynamic convolution and channel attention Journal: International Journal of Computing Science and Mathematics (IJCSM) 2025 Vol.21 No.2 pp.151 - 161 Abstract: 3D object detection as a very important part of the autonomous driving perception obtained a rapid development in recent years, but the existing 3D detector depends on the fixed weight nuclear convolution to deal with the information area, and then aggregate context to detect, these methods in the relationship between the capture point, lose some information. The dynamic selection process is more suitable for sparse and unordered point clouds data and can better extract features and aggregate context information. We propose the dynamic convolution and channel attention module (DCCA) to enhance the existing 3D detector. The attentional mechanism can be regarded as a dynamic selection process, which is realised by adaptive weights of features according to the importance of inputs. We use dynamic convolutional and channel attention mechanisms to enhance features. Achieving 81.7% of the results on the KITTI test 3D object detection benchmark at medium difficulty, and our method can be flexibly applied to most of the advanced detectors, improving the accuracy. Inderscience Publishers - linking academia, business and industry through research

Title: DCCA: 3D object detection with dynamic convolution and channel attention

Authors: Chen Jiang; Shuxia Lu; Tingting Ma; Xianghu Zhou

Addresses: Key Laboratory of Machine Learning and Computational Intelligence of Hebei Province, College of Mathematics and Information Science, Hebei University, Baoding, 071002, China ' Key Laboratory of Machine Learning and Computational Intelligence of Hebei Province, College of Mathematics and Information Science, Hebei University, Baoding, 071002, China ' Key Laboratory of Machine Learning and Computational Intelligence of Hebei Province, College of Mathematics and Information Science, Hebei University, Baoding, 071002, China ' Key Laboratory of Machine Learning and Computational Intelligence of Hebei Province, College of Mathematics and Information Science, Hebei University, Baoding, 071002, China

Abstract: 3D object detection as a very important part of the autonomous driving perception obtained a rapid development in recent years, but the existing 3D detector depends on the fixed weight nuclear convolution to deal with the information area, and then aggregate context to detect, these methods in the relationship between the capture point, lose some information. The dynamic selection process is more suitable for sparse and unordered point clouds data and can better extract features and aggregate context information. We propose the dynamic convolution and channel attention module (DCCA) to enhance the existing 3D detector. The attentional mechanism can be regarded as a dynamic selection process, which is realised by adaptive weights of features according to the importance of inputs. We use dynamic convolutional and channel attention mechanisms to enhance features. Achieving 81.7% of the results on the KITTI test 3D object detection benchmark at medium difficulty, and our method can be flexibly applied to most of the advanced detectors, improving the accuracy.

Keywords: autonomous vehicle; dynamic convolution; channel attention; LIDAR; 3D object detection.

DOI: 10.1504/IJCSM.2025.147113

International Journal of Computing Science and Mathematics, 2025 Vol.21 No.2, pp.151 - 161

Accepted: 27 Feb 2025
Published online: 10 Jul 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: DCCA: 3D object detection with dynamic convolution and channel attention

Keep up-to-date