Researchers have developed a new dataset and evaluation framework for object detection in autonomous driving systems specifically for Chinese rural environments. The study mixes real-world data captured in Weishi County, Henan, with synthetic data generated using Unreal Engine, creating a comprehensive 14-category object system that includes region-specific items like electric tricycles and roadside stalls. Experiments with 13 mainstream detectors, including various YOLO models and RT-DETR-L, showed that a moderate ratio of synthetic data (1:0.5) improved performance, with YOLO11m achieving the highest [email protected]. However, excessive synthetic data (1:1) led to domain shifts and performance degradation, particularly for identifying less common objects. AI
IMPACT This research provides empirical evidence for optimizing synthetic data strategies in autonomous driving, potentially accelerating deployment in under-represented rural environments.
RANK_REASON Academic paper detailing a new dataset and experimental evaluation of object detection models. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
