Advances, Systems and Applications
Table 1 Results of mAP evaluation against adversarial attack methods on the PASCAL VOC 2007 test set
From: RPU-PVB: robust object detection based on a unified metric perspective with bilinear interpolation
Method | Conference | Clean | \(A_{cls}\) [21] | \(A_{loc}\) [21] | CWA [19] | DAG [34] |
|---|---|---|---|---|---|---|
SSD [29] | ECCV2016 | 77.5 | 1.8 | 4.5 | 1.2 | 4.9 |
SSD-AT(\(A_{cls}\)) [29] | ICCV2019 | 46.7-30.8 | 21.8+20.0 | 32.2+27.7 | - | 28.0+23.1 |
SSD-AT(\(A_{loc}\)) [29] | ICCV2019 | 51.9-25.6 | 23.7+21.8 | 26.5+22.0 | - | 17.2+12.3 |
MTD [18] | ICCV2019 | 48.0-29.5 | 29.1+28.1 | 31.9+27.4 | 18.2+17.0 | 28.5+23.6 |
CWAT(PGD-10) [19] | CVPR2021 | 51.3-26.2 | 22.4+20.6 | 36.7+32.2 | 19.9+18.7 | 50.3+45.4 |
RobustDet [20] | ECCV2022 | 75.4-2.1 | 41.5+40.7 | 45.2+40.7 | 42.4+41.2 | 52.0 +47.1 |
RobustDet* [20] | ECCV2022 | 74.8-2.7 | 45.9+44.1 | 49.1+44.6 | 48.0+46.8 | 56.6+51.7 |
RPU-PVB | - | 73.0-4.5 | 60.2 +58.4 | 57.5 +53.0 | 60.9 +59.7 | 63.2 +58.3 |