Related Experiment Video
Updated: Jan 7, 2026
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
990
DDRN: DETR with dual refinement networks for autonomous vehicle object detection
Jiayao Li1, Chak Fong Cheang2,3,4, Zhaolong Du1
1School of Computer Science and Engineering, Faculty of Innovation Engineering, Macau University of Science and Technology, Taipa, Macao Special Administrative Region, China.
Scientific Reports
|December 24, 2025
Summary
Detecting small objects in autonomous driving is challenging. A new method, DETR with Dual Refinement Networks (DDRN), improves object detection accuracy and robustness for safer self-driving systems.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Robotics
Background:
- Object detection is critical for autonomous driving perception, decision-making, and control.
- Existing detectors struggle with small object detection and complex driving environments, impacting safety and efficiency.
Purpose of the Study:
- To enhance object detection performance in autonomous driving systems.
- To address the challenges of small object detection and environmental complexity.
Main Methods:
- Proposing Feature Pyramid Network-α (FPN-α) to improve small object detection by reweighting deep layer participation.
- Introducing Classification Refinement Networks (CRN) and Localization Refinement Networks (LRN) for optimized classification and localization.
- Integrating these components into DETR with Dual Refinement Networks (DDRN).
Main Results:
- Achieved 1.4% and 1.1% average precision (AP) improvements on public autonomous driving datasets.
- Demonstrated superior performance compared to several advanced object detectors.
- Validated the effectiveness, robustness, and applicability of the proposed DDRN method.
Conclusions:
- DDRN effectively addresses limitations in small object detection and complex environments.
- The proposed method significantly enhances the safety and efficiency of autonomous driving perception systems.
- DDRN shows strong potential for real-world autonomous driving applications.
