CADENet: Condition-Adaptive Asynchronous Dual-Stream Enhancement Network for Adverse Weather Perception in Autonomous Driving

2026-05-19

AI Summary

This work addresses the significant performance degradation of camera-based object detection under adverse weather conditions, a problem exacerbated by latency and annotation incompleteness bias in existing enhancement-then-detection pipelines. The authors propose a training-free, three-thread asynchronous system: the main thread performs real-time detection at full frame rate; an enhancement thread adaptively optimizes image quality and fuses detection results; and a third thread leverages zero-shot CLIP classification to identify weather types, enabling plug-and-play weather adaptation. By integrating Conditionally Adaptive Enhancement (CAPE) with Entropy-Guided Non-Maximum Suppression (EG-NMS), the method improves robustness without introducing additional latency and formally analyzes annotation bias to isolate true performance gains. Evaluated on the DAWN dataset, it achieves F1 scores of 0.0230 in snow and 0.0038 in rain (micro Recall: 0.0103), while maintaining a stable 44 FPS on the main thread—without retraining or auxiliary sensors.

Abstract

Adverse weather (rain, fog, sand, and snow) degrades camera-based object detection in autonomous vehicles. Existing enhancement-then-detect approaches stall the safety-critical perception loop, violating hard real-time requirements. Progress on this problem is also constrained by an under-recognized evaluation ceiling: ground truth annotated on degraded images cannot credit a detector that recovers objects the annotators themselves could not see, so a genuinely useful enhancement can register as a near-flat F1 gain. This paper presents CADENet (Condition-Adaptive Asynchronous Dual-stream Enhancement Network), a training-free three-thread system: Thread S (YOLOv11n) delivers detections at full frame rate with zero added latency; Thread Q applies condition-adaptive enhancement (CAPE) and fuses results via entropy-guided NMS (EG-NMS) without blocking Thread S; Thread E provides CLIP zero-shot weather classification, so new weather categories require only a new text prompt, with no labeled data and no retraining. Evaluated on 1327 DAWN images (YOLOv11m, IoU = 0.5, confidence = 0.25), CADENet achieves Recall = 0.0103 (micro), F1 = 0.0230 on snow, and F1 = 0.0038 on rain. We formalize the annotation completeness bias on DAWN-class data, so the reported F1 values are lower bounds on the true gain; recall is the annotation-gap-immune headline metric. Thread S sustains approximately 44 FPS regardless of enhancement load. No model retraining or additional sensor hardware is required.

Problem

• adverse weather perception
• real-time object detection
• annotation completeness bias
• autonomous driving
• image enhancement

Innovation

• asynchronous dual-stream
• condition-adaptive enhancement
• zero-shot weather classification
• entropy-guided NMS
• annotation completeness bias