From: Multitask adversarial attack with dispersion amplification
Base model | Attack | Det. YoloV3 mAP | Det. Mask-RCNN mAP | Det. Faster-RCNN mAP | Seg. DeepLab mAP | Classification | PSNR |
---|---|---|---|---|---|---|---|
VGG16 | FGSM | 0.541 | 0.506 | 0.522 | 0.599 | 0.483 | 30.01 |
BIM | 0.522 | 0.517 | 0.512 | 0.632 | 0.349 | 25.86 | |
LBFGS | – | 0.542 | 0.49 | 0.538 | – | 25.51 | |
DR | 0.126 | 0.586 | 0.136 | 0.411 | 0.112 | 30.96 | |
DA | 0.119 | 0.144 | 0.136 | 0.128 | 0.128 | 28.32 | |
ResNet101 | FGSM | 0.612 | 0.578 | 0.607 | 0.622 | 0.381 | 30.01 |
BIM | 0.628 | 0.571 | 0.567 | 0.585 | 0.269 | 26.68 | |
LBFGS | – | 0.56 | 0.553 | 0.547 | – | 26.73 | |
DR | 0.138 | 0.129 | 0.131 | 0.530 | 0.116 | 32.69 | |
DA | 0.105 | 0.112 | 0.112 | 0.174 | 0.060 | 30.33 | |
InceptionV3 | FGSM | 0.595 | 0.613 | 0.585 | 0.594 | 0.520 | 30.01 |
BIM | 0.601 | 0.609 | 0.611 | 0.591 | 0.448 | 26.81 | |
LBFGS | – | 0.515 | 0.604 | 0.549 | – | 26.56 | |
DR | 0.132 | 0.130 | 0.111 | 0.490 | 0.074 | 33.31 | |
DA | 0.126 | 0.112 | 0.110 | 0.216 | 0.032 | 30.68 | |
MobileNet v2 | FGSM | 0.622 | 0.641 | 0.629 | 0.672 | 0.325 | 30.01 |
BIM | 0.570 | 0.595 | 0.601 | 0.639 | 0.338 | 26.38 | |
LBFGS | – | 0.553 | 0.575 | 0.539 | – | 25.83 | |
DR | 0.147 | 0.147 | 0.133 | 0.221 | 0.104 | 30.55 | |
DA | 0.112 | 0.109 | 0.125 | 0.147 | 0.092 | 29.98 |