Interactive demonstration of domain-generalized pothole detection with enhancement-aware approach. Upload road images to detect potholes automatically.
Drag and drop your image here
or click to browse (JPG, PNG)Upload an image to see detection results
Processing image with enhancement module...
Select a condition to browse images
Based on Bučko et al. (2022), pothole detection accuracy degrades up to 90.6% at night (0.747 → 0.0701 mAP). Our enhancement-aware approach recovers accuracy by 401% on night conditions.
Condition-aware enhancement pipeline: Clear Rain Sunset Night
Road Damage Dataset with 1,836 images across 5 conditions. Training includes domain randomization and augmentation for better generalization.
YOLO11-n achieves ~180 FPS on edge devices with competitive accuracy. YOLO11-s provides better accuracy at ~140 FPS. Both support real-time applications.