Related Experiment Video
Updated: Sep 16, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
644
CIDRA-Net: Cross-modal interaction fusion network with distribution-relation awareness for robust 3D object
Hongqi Yu1, Xiaoqin Zhang1, Xiaolong Zhou2
1Key Laboratory of Intelligent Informatics for Safety & Emergency of Zhejiang Province, Wenzhou University, Wenzhou, 325035, Zhejiang, China.
Summary
CIDRA-Net enhances 3D object detection for autonomous driving by fusing LiDAR and camera data. This novel approach improves object classification and localization by considering distribution and relation awareness.
Area of Science:
- Computer Vision
- Robotics
- Artificial Intelligence
Background:
- 3D object detection is vital for autonomous driving systems.
- Current methods struggle with fusing multi-modal data (LiDAR, camera) effectively.
- Challenges include point cloud sparsity and complex object relationships.
Purpose of the Study:
- To propose CIDRA-Net, a novel cross-modal fusion network for 3D object detection.
- To enhance the learning of camera semantics and LiDAR spatial information.
- To address distribution imbalances and implicit relational contexts in 3D object detection.
Main Methods:
- Region Cross-Modal Interaction Fusion (RCIF) using dual-modal attention.
- Dual-Branch Distribution Perception (DBDP) for learning point distributions.
- Global-Local Relation Mining (GLRM) for contextual information capture.
Main Results:
- Achieved state-of-the-art performance on nuScenes and KITTI benchmarks.
- Demonstrated strong generalization across different backbones.
- Showcased robustness against sensor errors.
Conclusions:
- CIDRA-Net effectively fuses multi-modal data for superior 3D object detection.
- The proposed modules improve understanding of object distributions and relations.
- The network offers a robust and generalizable solution for autonomous driving perception.
