Table 5.
Comparison of Average Precision (AP) on GC10-DET dataset. The proposed method performs the highest AP values for three defect categories. The proposed method also provides the highest mAP value. The bold helps to emphasize the highest data.
Types | AP | ||||
---|---|---|---|---|---|
SSD | Faster-RCNN | YOLO-V2 | YOLO-V3 | Proposed Method | |
Pu | 0.860 | 0.899 | 0.725 | 0.836 | 0.900 |
Wl | 0.974 | 0.554 | 0.328 | 0.241 | 0.885 |
Cg | 0.861 | 0.872 | 0.819 | 0.752 | 0.848 |
Ws | 0.552 | 0.599 | 0.476 | 0.495 | 0.558 |
Os | 0.612 | 0.653 | 0.403 | 0.329 | 0.622 |
Ss | 0.689 | 0.579 | 0.473 | 0.325 | 0.650 |
In | 0.168 | 0.194 | 0.096 | 0.036 | 0.256 |
Rp | 0.105 | 0.364 | 0.018 | 0.036 | 0.364 |
Cr | 0.527 | 0.736 | 0.212 | 0.429 | 0.521 |
Wf | 1.000 | 0.818 | 0.614 | 0.400 | 0.919 |
mAP | 0.635 | 0.627 | 0.433 | 0.388 | 0.651 |