-
Notifications
You must be signed in to change notification settings - Fork 0
/
gtsrb.txt
35 lines (32 loc) · 2.31 KB
/
gtsrb.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
FGSM:
Robust accuracy of original adversarial attack: 30.898437499999996%
Robust accuracy of modified adversarial attack: 67.265625%
Robust accuracy of reconstructed adversarial attack: 29.6875%
Setting up [LPIPS] perceptual loss: trunk [alex], v[0.1], spatial [off]
Loading model from: /scratch/itee/uqsswain/miniconda3/envs/py38/lib/python3.8/site-packages/lpips/weights/v0.1/alex.pth
Average LPIPS score of original adversarial attack: tensor(0.0838, grad_fn=<MeanBackward0>)
Average LPIPS score of modifed adversarial attack: tensor(0.0690, grad_fn=<MeanBackward0>)
PGD
Robust accuracy of original adversarial attack: 17.08984375%
Robust accuracy of modified adversarial attack: 32.7734375%
Robust accuracy of reconstructed adversarial attack: 2.40234375%
Setting up [LPIPS] perceptual loss: trunk [alex], v[0.1], spatial [off]
Loading model from: /scratch/itee/uqsswain/miniconda3/envs/py38/lib/python3.8/site-packages/lpips/weights/v0.1/alex.pth
Average LPIPS score of original adversarial attack: tensor(0.0824, device='cuda:0', grad_fn=<MeanBackward0>)
Average LPIPS score of modifed adversarial attack: tensor(0.0737, device='cuda:0', grad_fn=<MeanBackward0>)
CnW
Robust accuracy of original adversarial attack: 19.47265625%
Robust accuracy of modified adversarial attack: 91.15234375%
Robust accuracy of reconstructed adversarial attack: 5.95703125%
Setting up [LPIPS] perceptual loss: trunk [alex], v[0.1], spatial [off]
Loading model from: /scratch/itee/uqsswain/miniconda3/envs/py38/lib/python3.8/site-packages/lpips/weights/v0.1/alex.pth
Average LPIPS score of original adversarial attack: tensor(0.1270, device='cuda:0', grad_fn=<MeanBackward0>)
Average LPIPS score of modifed adversarial attack: tensor(0.0684, device='cuda:0', grad_fn=<MeanBackward0>)
EAD
Robust accuracy of original adversarial attack: 10.4296875%
Robust accuracy of modified adversarial attack: 68.6328125%
Robust accuracy of reconstructed adversarial attack: 0.0%
Setting up [LPIPS] perceptual loss: trunk [alex], v[0.1], spatial [off]
Loading model from: /scratch/itee/uqsswain/miniconda3/envs/py38/lib/python3.8/site-packages/lpips/weights/v0.1/alex.pth
Average LPIPS score of original adversarial attack: tensor(0.2658, device='cuda:0', grad_fn=<MeanBackward0>)
Average LPIPS score of modifed adversarial attack: tensor(0.0934, device='cuda:0', grad_fn=<MeanBackward0>)