Table 1 Performance comparison in terms of mAP (%) between our proposed method and the state-of-the-art algorithms on CLD dataset, NEU steel surface defect dataset, and PKU-Market-PCB dataset, where PvTv2 means PVTv2-B2-Li. Except FPFM, We re-implemented baselines on three databases.
From: Sparse cross-transformer network for surface defect detection
Method | Backbone | CLD dataset | |||||
|---|---|---|---|---|---|---|---|
1-shot | 2-shot | 3-shot | 5-shot | 10-shot | 30-shot | ||
MetaRCNN | ResNet101 | 10.86 | 18.25 | 23.21 | 27.55 | 31.73 | 52.42 |
TFA | ResNet101 | 12.97 | 15.70 | 19.50 | 26.47 | 31.89 | 51.73 |
FSCE | ResNet101 | 15.28 | 21.32 | 23.00 | 29.37 | 34.19 | 52.61 |
AttentionRPN | ResNet50 | 16.34 | 22.50 | 23.69 | 25.97 | 31.49 | 49.86 |
FSDetView | ResNet101 | 14.11 | 21.90 | 22.80 | 25.56 | 30.87 | 47.90 |
MPSR | ResNet101 | 21.26 | 27.63 | 30.37 | 38.76 | 42.8 | 55.91 |
QA-FewDet | ResNet101 | 16.57 | 22.24 | 25.68 | 31.28 | 37.19 | 51.64 |
DeFRCN | ResNet101 | 18.22 | 26.81 | 32.60 | 39.73 | 46.26 | 58.23 |
FCT | PvTv2 | 25.16 | 28.64 | 32.34 | 40.32 | 45.62 | 59.81 |
Ours | PvTv2 | 26.53 | 31.92 | 34.98 | 41.9 | 47.21 | 62.73 |
Method | Backbone | NEU steel surface defect dataset | |||||
|---|---|---|---|---|---|---|---|
1-shot | 2-shot | 3-shot | 5-shot | 10-shot | 30-shot | ||
MetaRCNN | ResNet101 | 12.21 | 18.07 | 26.3 | 43.12 | 52.06 | 66.17 |
TFA | ResNet101 | 11.93 | 14.99 | 26.92 | 45.47 | 52.31 | 65.67 |
FSCE | ResNet101 | 16.19 | 26.97 | 33.87 | 46.17 | 60.13 | 72.08 |
AttentionRPN | ResNet50 | 18.83 | 26.33 | 32.18 | 47.07 | 62.53 | 73.81 |
FSDetView | ResNet101 | 16.12 | 21.08 | 31.37 | 43.19 | 64.28 | 75.97 |
MPSR | ResNet101 | 19.49 | 25.99 | 39.76 | 49.53 | 66.87 | 78.29 |
QA-FewDet | ResNet101 | 15.91 | 18.09 | 32.19 | 43.37 | 62.12 | 74.01 |
DeFRCN | ResNet101 | 21.06 | 26.20 | 39.93 | 48.95 | 67.37 | 80.46 |
FCT | PvTv2 | 29.07 | 33.18 | 43.71 | 56.64 | 70.16 | 83.58 |
Ours | PvTv2 | 29.81 | 35.26 | 46.49 | 58.94 | 72.18 | 85.29 |
Method | Backbone | PKU-Market-PCB dataset | |||||
|---|---|---|---|---|---|---|---|
1-shot | 2-shot | 3-shot | 5-shot | 10-shot | 30-shot | ||
MetaRCNN | ResNet101 | 8.74 | 14.53 | 21.85 | 44.78 | 57.83 | 69.34 |
TFA | ResNet101 | 10.21 | 14.68 | 22.72 | 43.69 | 55.18 | 68.51 |
FSCE | ResNet101 | 17.58 | 28.16 | 36.52 | 43.98 | 63.41 | 75.50 |
AttentionRPN | ResNet50 | 19.05 | 24.83 | 31.61 | 50.29 | 65.47 | 76.50 |
FSDetView | ResNet101 | 11.92 | 18.12 | 30.94 | 40.52 | 62.19 | 73.41 |
MPSR | ResNet101 | 12.59 | 17.25 | 28.32 | 45.76 | 59.45 | 71.86 |
QA-FewDet | ResNet101 | 13.20 | 20.40 | 32.30 | 49.80 | 63.9 | 75.97 |
DeFRCN | ResNet101 | 18.70 | 26.10 | 39.50 | 45.10 | 68.20 | 83.28 |
FCT | PvTv2 | 26.29 | 30.21 | 42.95 | 55.32 | 69.86 | 82.60 |
FPFM | - | 16.26 | 22.07 | 39.36 | 49.81 | 69.52 | 78.86 |
Ours | PvTv2 | 28.20 | 36.80 | 48.30 | 59.10 | 73.60 | 88.70 |
