Table 7 Performance comparison of each method on NEU-DET datasets.
From: PEYOLO a perception efficient network for multiscale surface defects detection
Method | P↑ | R↑ | mAP50↑ | Params↓ | FPS↑ |
|---|---|---|---|---|---|
YOLOv3-tiny | 0.571 | 0.679 | 67.7 | 12.17 | 226.8 |
YOLOv4-csp | 0.663 | 0.682 | 72.6 | 59.58 | 39.4 |
YOLOv5n | 0.637 | 0.714 | 72.9 | 2.65 | 184.1 |
YOLOv6n | 0.681 | 0.705 | 73.8 | 4.50 | 226.4 |
YOLOv8n | 0.668 | 0.71 | 74.6 | 3.15 | 198.1 |
YOLOv8-ghost | 0.672 | 0.706 | 76.4 | 1.86 | 157.5 |
YOLOv8-world | 0.662 | 0.726 | 76.8 | 4.20 | 124.6 |
YOLOv8-worldv2 | 0.697 | 0.726 | 75.2 | 3.69 | 139.5 |
YOLOv9c | 0.616 | 0.741 | 73.0 | 25.59 | 51.5 |
YOLOv10n | 0.634 | 0.639 | 69.4 | 2.77 | 136.2 |
Faster-rcnn (Backbone: resnet50) | 0.242 | 0.572 | 38.6 | 136.79 | 18.6 |
SSD (Backbone: resnet50) | 0.781 | 0.093 | 36.1 | 12.33 | 53.6 |
EfficientDet-d1 | 0.643 | 0.261 | 44.8 | 6.55 | 19.9 |
FCOS | 0.781 | 0.551 | 70.7 | 32.13 | 35.6 |
PEYOLO | 0.709 | 0.745 | 78.1 | 3.09 | 125.4 |
