Error in TTAA code

[nano-vision]

Hi, I met some errors when running TTAA code. Could you please re-check your code? Thanks a lot.

1. gaussian_smoothing is missing

AI-Security/Adversarial Example/TTAA/train.py

Line 17 in 7294840

from gaussian_smoothing import *

2. there are a lot of variables which are not defined before in train.py.

feature_layer in Line 146, Line 156, and Line 170

feature in Line157

AI-Security/Adversarial Example/TTAA/train.py

Lines 145 to 157 in 7294840

	_, t_features = l_d(normalize(t_img.clone().detach()))
	t_feature = t_features[feature_layer]
	t_feature_similarity = f_d(t_feature)
	adv = netG(img.clone())
	# Projection
	adv = torch.min(torch.max(adv, img - eps), img + eps)
	adv = torch.clamp(adv, 0.0, 1.0)
	_, s_features = l_d(normalize(adv.clone().detach()))
	s_feature = s_features[feature_layer]
	s_feature_similarity = f_d(feature)

t_outputs in Line 161

AI-Security/Adversarial Example/TTAA/train.py

Line 161 in 7294840

D_real_loss = BCE(t_feature_similarity, torch.ones(t_outputs.shape).detach().to(device))

[Farhamdur]

@stark-cs Have you been able to run the TTAA code?

[nano-vision]

@stark-cs Have you been able to run the TTAA code?

No