v1.1

.gitignore

@@ -8,3 +8,4 @@ demos/data/cifar*
 MENIFEST.in
 setup.cfg
 setup.py
+data/

README.md

@@ -129,4 +129,8 @@ This demo shows how to do adversarial training with this repository. The MNIST d
 ### Version 1.0
 * **attack.py** :
     * **load** : Load is depreciated. Instead, use TensorDataset and DataLoader.
-    * **save** : The problem of calculating invalid accuracy when the mode of the attack set to 'int' is solved.
+    * **save** : The problem of calculating invalid accuracy when the mode of the attack set to 'int' is solved.
+
+### Version 1.1
+* **DeepFool** :
+    * [**Error solved**](https://github.com/Harry24k/adversairal-attacks-pytorch/issues/2).

torchattacks/attacks/deepfool.py

@@ -40,8 +40,9 @@ def forward(self, images, labels):
                 image.requires_grad = True
                 output = self.model(image)[0]
                 _, pre = torch.max(output, 0)
-
+            
                 if pre != pre_0 :
+                    image = torch.clamp(image, min=0, max=1).detach()
                     break
 
                 r = None