ARTFEED — Contemporary Art Intelligence

CADENet: Asynchronous Enhancement for Adverse Weather Detection

ai-technology · 2026-05-20

A team of researchers has introduced CADENet, an innovative system designed for camera-based object detection in autonomous vehicles operating under challenging weather conditions like rain, fog, sand, and snow, without the need for training. The system consists of three threads: Thread S utilizes YOLOv11n at full frame rate with no additional latency; Thread Q implements condition-adaptive enhancement (CAPE) and combines results through entropy-guided NMS, allowing Thread S to continue uninterrupted; and Thread E employs CLIP for weather classification without prior training. CADENet aims to overcome the limitations of evaluating detectors on degraded images and is tailored to fulfill stringent real-time demands in safety-critical perception scenarios.

Key facts

  • CADENet is a training-free three-thread system
  • Thread S uses YOLOv11n for detections at full frame rate
  • Thread Q applies CAPE enhancement and EG-NMS fusion
  • Thread E uses CLIP for zero-shot weather classification
  • Addresses adverse weather: rain, fog, sand, snow
  • Tackles evaluation ceiling in degraded image detection
  • Aims to meet hard real-time requirements
  • Proposed for autonomous driving perception

Entities

Sources