Research Article
Pothole Detection Using Deep Learning: A Real-Time and AI-on-the-Edge Perspective
Table 1
Performance evaluation of each model on test subset of pothole image dataset (PID)
| Model | Precision | Recall | F1-score | [email protected] (%) | Inference time (ms) |
| SSD-Mobilenetv2 | 0.42 | 0.56 | 0.479 | 47.4 | 7 | YOLOv1 | 0.82 | 0.69 | 0.74 | 79.55 | 340 | YOLOv2 | 0.81 | 0.76 | 0.78 | 81.21 | 33.7 | YOLOv3 | 0.77 | 0.78 | 0.78 | 83.60 | 70.57 | Tiny-YOLOv4 | 0.76 | 0.75 | 0.76 | 80.04 | 4.86 | YOLOv4 | 0.81 | 0.83 | 0.82 | 85.48 | 52.51 | YOLOv5 | 0.93 | 0.83 | 0.87 | 95.00 | 10 |
|
|