| vgg19 |
tageted cost + non tageted cost，一比一，迭代10次 |
0.082 |
基本识别错误，但没有识别到taget类 |
| vgg19 |
tageted cost，迭代20次 |
0.079 |
大部分能识别到taget类 |
| resnet50 |
|
|
没加噪声的原图识别率极低，暂时不明原因 |
| vgg19+inception_v3 |
tageted cost + non tageted cost，一比一，先vgg19迭代20次，再inceptionV3迭代20次 |
0.101 |
基本识别错误，但没有识别到taget类 |
| inception_v3 |
先用IFGSM迭代10次，然后ILA迭代10次 |
0.156 |
迁移到resnet18上测试，分类准确率30%最左右 |
| resnet18 |
先用IFGSM迭代10次，然后ILA迭代10次 |
0.100 |
迁移到InceptionV3上测试，分类准确率到80% |
| Densenet121 |
同上 |
0.1339 |
迁移到InceptionV3上测试，分类准确率76%，而仅使用IFGSM的方法，分类准确率为93% |
| Alexnet |
同上 |
0.1102 |
迁移到InceptionV3上测试，分类准确率91%，而仅使用IFGSM的方法，分类准确率为93% |
| SqueezeNet1.0 L3 |
同上 |
0.1200 |
迁移到InceptionV3上测试，分类准确率74%，而仅使用IFGSM的方法，分类准确率为90% |
| inception_v3, inception_v4, inception_resnet_v2, resnet_V2, prob=0.7, momentum=1.0, num_iter=16 |
Translation-Invariant-Attacks, untargeted |
0.49 |
|
| inception_v3, inception_v4, inception_resnet_v2, resnet_V2, prob=0.7, momentum=1.0, num_iter=16 |
Translation-Invariant-Attacks, targeted |
0.44 |
|
| inception_v3, inception_v4, inception_resnet_v2, resnet_V2, prob=0.5, momentum=1.0, num_iter=50 |
Translation-Invariant-Attacks, untargeted |
0.66 |
|
| adv_inception_v3, ens_adv_inception_resnet_v2, nception_v3, inception_v4, inception_resnet_v2, resnet_V2, prob=0.7, momentum=1.0, num_iter=10 |
Translation-Invariant-Attacks, targeted |
0.48 |
|
| adv_inception_v3, ens_adv_inception_resnet_v2, nception_v3, inception_v4, inception_resnet_v2, resnet_V2, prob=0.7, momentum=1.0, num_iter=50 |
Translation-Invariant-Attacks, targeted |
0.51 |
|
| adv_inception_v3, ens3_adv_inception_v3, ens3_adv_inception_v3, ens_adv_inception_resnet_v2, inception_v4, adv_inception_resnet_v2, resnet_V2, prob=0.0, momentum=1.0, num_iter=50 |
Translation-Invariant-Attacks, untargeted |
1.02 |
|