From eec72da9c1ebd87adf0a9f830d60f5c57d3d6a29 Mon Sep 17 00:00:00 2001
From: Nataniel Ruiz <nruiz@Nataniels-MacBook-Pro.local>
Date: Wed, 25 Dec 2019 20:42:07 -0400
Subject: [PATCH] next

---
 cyclegan/models/test_model.py | 5 -----
 cyclegan/test.py              | 7 -------
 cyclegan/util/attacks.py      | 4 +---
 3 files changed, 1 insertion(+), 15 deletions(-)

diff --git a/cyclegan/models/test_model.py b/cyclegan/models/test_model.py
index d7eaa8d..c71f708 100644
--- a/cyclegan/models/test_model.py
+++ b/cyclegan/models/test_model.py
@@ -78,8 +78,6 @@ class TestModel(BaseModel):
         black = np.zeros((1, 3, image.size(2), image.size(3)))
         black = torch.FloatTensor(black).cuda()
         input_adv, perturb = pgd_attack.perturb(image, black)      
-
-        black = None
         
         return input_adv, perturb
 
@@ -94,9 +92,6 @@ class TestModel(BaseModel):
         l2 = F.mse_loss(generated, generated_noattack)
         l0 = (generated - generated_noattack).norm(0)
         d = (generated - generated_noattack).norm(float('-inf'))
-
-        generated = None
-        generated_noattack = None
         
         return l1, l2, l0, d
 
diff --git a/cyclegan/test.py b/cyclegan/test.py
index 8ce44af..2e49a6e 100644
--- a/cyclegan/test.py
+++ b/cyclegan/test.py
@@ -85,13 +85,6 @@ if __name__ == '__main__':
             print('processing (%04d)-th image... %s' % (i, img_path))
         save_images(webpage, visuals, img_path, aspect_ratio=opt.aspect_ratio, width=opt.display_winsize)
 
-        model.fake = None
-        model.real = None
-        model.fake_noattack = None
-        data = None
-        perturb = None
-        input_adv = None
-
     webpage.save()  # save the HTML
 
 
diff --git a/cyclegan/util/attacks.py b/cyclegan/util/attacks.py
index b6296ff..18b90cd 100644
--- a/cyclegan/util/attacks.py
+++ b/cyclegan/util/attacks.py
@@ -35,10 +35,8 @@ class LinfPGDAttack(object):
 
             eta = torch.clamp(X_adv - X_nat, min=-self.epsilon, max=self.epsilon)
             X = torch.clamp(X_nat + eta, min=-1, max=1).detach_()
-            eta = None
-            X_adv = None
 
-        return X, X - X_nat
+        return X, eta
 
 def clip_tensor(X, Y, Z):
     # Clip X with Y min and Z max