Table 6 Comparison with the SOTA semi-supervised segmentation methods in the SynLCD dataset across varying proportions of labeled data (from 0% to 15%). All models are pre-trained on the abpt defects and subsequently fine-tuned and tested using the line defects. The bold font indicates the best results.
From: Siamese network with change awareness for surface defect segmentation in complex backgrounds
Method | \(\hbox {mIoU}\uparrow\) | \(\hbox {Fscore}\uparrow\) | ||||||
|---|---|---|---|---|---|---|---|---|
0% | 5% | 10% | 15% | 0% | 5% | 10% | 15% | |
DCT45 | 0.05 | 56.96 | 73.67 | 71.85 | 0.10 | 71.27 | 84.57 | 82.75 |
UAMT46 | 0.44 | 61.68 | 68.73 | 71.96 | 0.88 | 75.48 | 80.94 | 83.15 |
CPS43 | 1.09 | 65.07 | 65.63 | 76.02 | 2.15 | 78.29 | 78.70 | 85.68 |
UCC44 | 0.015 | 61.40 | 70.48 | 71.55 | 0.03 | 75.41 | 82.27 | 82.78 |
UAPS5 | 0.44 | 58.86 | 74.43 | 81.34 | 0.88 | 72.52 | 84.35 | 89.22 |
Our-CADNet | 46.89 | 82.93 | 84.52 | 84.71 | 63.84 | 90.87 | 91.64 | 91.72 |
