Table 5 Comparison with the mainstream semantic segmentation methods in the MvTec-AD Dataset. Bold, Bolditalic and italic indicate the top three results. Note that there are 15 classes in MvTec-AD and six of them are reported here.
From: Siamese network with change awareness for surface defect segmentation in complex backgrounds
Method | \(\hbox {IOU}_{c1}\uparrow\) | \(\hbox {IOU}_{c2}\uparrow\) | \(\hbox {IOU}_{c3}\uparrow\) | \(\hbox {IOU}_{c4}\uparrow\) | \(\hbox {IOU}_{c5} \uparrow\) | \(\hbox {IOU}_{c6}\uparrow\) | \(\hbox {mIOU}\uparrow\) | \(\hbox {mAcc}\uparrow\) | \(\hbox {mFscore}\uparrow\) |
|---|---|---|---|---|---|---|---|---|---|
FCN66 | 76.10 | 60.14 | 35.93 | 69.73 | 13.51 | 79.65 | 58.14 | 64.84 | 70.00 |
PSPNet67 | 72.00 | 68.24 | 43.86 | 74.89 | 42.43 | 83.44 | 65.42 | 76.25 | 77.58 |
DeepLabV3+36 | 76.65 | 63.48 | 41.18 | 72.31 | 34.93 | 81.12 | 63.77 | 77.59 | 76.19 |
DANet68 | 75.13 | 56.37 | 37.95 | 72.42 | 27.10 | 80.92 | 61.63 | 72.49 | 73.94 |
OCRNet69 | 70.89 | 65.18 | 45.67 | 65.47 | 35.41 | 81.51 | 59.89 | 68.98 | 72.31 |
SegFormer40 | 81.63 | 64.63 | 53.81 | 70.81 | 44.14 | 84.71 | 65.97 | 71.21 | 77.51 |
Our-CADNet | 82.60 | 74.16 | 61.19 | 73.06 | 52.69 | 86.41 | 71.35 | 80.85 | 82.24 |
