Hierarchically Decoupled Mixture-of-Experts for Robust Traffic Sign Recognition in Complex Driving Scenarios
Researchers have developed a new Mixture-of-Experts (MoE) framework called CBDES MoE TSR to improve traffic sign recognition in autonomous driving. This approach uses a dynamic routing mechanism that selects the most suitable expert model for a given image, moving away from static, globally shared parameters. The system demonstrated a 2.3% increase in mAP50-95 accuracy, reaching 76.8%, while simultaneously reducing computational overhead by 39.4% on a composite traffic sign dataset. AI
IMPACT Enhances robustness of autonomous driving perception systems in varied conditions.