Table 5 Accuracy changes for different algorithms on the PASCAL VOC dataset.
From: DAM-Faster RCNN: few-shot defect detection method for wood based on dual attention mechanism
Methods/shot | Novel-class split1 | Novel-class split2 | Novel-class split3 | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1 | 2 | 3 | 5 | 10 | 1 | 2 | 3 | 5 | 19 | 1 | 2 | 3 | 5 | 10 | |
Meta Det26 | 18.7 | 20.7 | 30.5 | 36.5 | 49.6 | 21.6 | 22.8 | 27.8 | 31.4 | 42.5 | 20.2 | 24.1 | 29.3 | 43.6 | 44.3 |
Meta RCNN27 | 17.9 | 26.7 | 34.5 | 43.7 | 53.7 | 9.7 | 20.4 | 30.4 | 35.1 | 46.2 | 15.7 | 19.3 | 28.6 | 42.7 | 48.8 |
TFA w/fc28 | 36.4 | 29.4 | 44.0 | 55.4 | 56.9 | 18.1 | 29.1 | 33.6 | 35.7 | 38.5 | 27.3 | 33.7 | 42.9 | 49.5 | 49.7 |
TFA w/cos28 | 36.9 | 36.5 | 45.1 | 55.9 | 56.2 | 24.0 | 26.8 | 32.3 | 34.8 | 39.2 | 30.7 | 35.2 | 42.9 | 49.5 | 49.7 |
FSCE29 | 38.9 | 41.9 | 52.3 | 53.4 | 58.7 | 27.9 | 31.7 | 39.3 | 33.2 | 47.8 | 33.6 | 34.7 | 39.5 | 50.7 | 53.7 |
Ours | 38.4 | 42.3 | 53.6 | 56.0 | 59.6 | 24.8 | 33.7 | 38.7 | 45.9 | 53.8 | 29.7 | 37.8 | 45.9 | 51.5 | 53.9 |
