diff --git a/adversarial/README.md b/adversarial/README.md
index 7c9502828f35a4f78bdf59f2e287b5e0c3fdd411..51da21918a9d6e2192a2e03eabef4fde97896bc5 100644
--- a/adversarial/README.md
+++ b/adversarial/README.md
@@ -1,3 +1,9 @@
 # Advbox
 
-Advbox is a Python toolbox to create adversarial examples that fool neural networks. It requires Python and paddle.
\ No newline at end of file
+Advbox is a Python toolbox to create adversarial examples that fool neural networks. It requires Python and paddle.
+
+## How to use
+
+1. train a model and save it's parameters. (like fluid_mnist.py)
+2. load the parameters which is trained in step1, then reconstruct the model.(like mnist_tutorial_fgsm.py)
+3. use advbox to generate the adversarial sample.
diff --git a/adversarial/advbox/__init__.py b/adversarial/advbox/__init__.py
index 4beb6be0a20d52e499c7da6263b5d8f4f105c9c6..f56f14f18dafdfe1e712cea178a63f09a087b587 100644
--- a/adversarial/advbox/__init__.py
+++ b/adversarial/advbox/__init__.py
@@ -11,7 +11,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
 """
    A set of tools for generating adversarial example on paddle platform 
 """
diff --git a/adversarial/advbox/attacks/base.py b/adversarial/advbox/attacks/base.py
index 9cc2bfb8543ee694c3b365eec9732e8dd984d475..dab1dbbeb0270d303dbec1bd0b28a43e9d26f075 100644
--- a/adversarial/advbox/attacks/base.py
+++ b/adversarial/advbox/attacks/base.py
@@ -7,6 +7,7 @@ import abc
 
 abstractmethod = abc.abstractmethod
 
+
 class Attack(object):
     """
     Abstract base class for adversarial attacks. `Attack` represent an adversarial attack
diff --git a/adversarial/advbox/attacks/gradientsign.py b/adversarial/advbox/attacks/gradientsign.py
index 6c188f6249830cb381f55c58e5978f7ad935805d..37fbdb11328a66a37ec58621ad92dbd3db6d58ed 100644
--- a/adversarial/advbox/attacks/gradientsign.py
+++ b/adversarial/advbox/attacks/gradientsign.py
@@ -5,7 +5,8 @@ from __future__ import division
 import numpy as np
 from collections import Iterable
 from .base import Attack
- 
+
+
 class GradientSignAttack(Attack):
     """
     This attack was originally implemented by Goodfellow et al. (2015) with the
@@ -22,10 +23,11 @@ class GradientSignAttack(Attack):
         gradient_sign = np.sign(gradient) * (max_ - min_)
 
         if not isinstance(epsilons, Iterable):
-            epsilons = np.linspace(0, 1, num = epsilons + 1)
+            epsilons = np.linspace(0, 1, num=epsilons + 1)
 
         for epsilon in epsilons:
-            adv_img = image_batch[0][0].reshape(gradient_sign.shape) + epsilon * gradient_sign
+            adv_img = image_batch[0][0].reshape(
+                gradient_sign.shape) + epsilon * gradient_sign
             adv_img = np.clip(adv_img, min_, max_)
             adv_label = np.argmax(self.model.predict([(adv_img, 0)]))
             #print("pre_label="+str(pre_label)+ " adv_label="+str(adv_label))
@@ -33,4 +35,5 @@ class GradientSignAttack(Attack):
                 #print(epsilon, pre_label, adv_label)
                 return adv_img
 
+
 FGSM = GradientSignAttack
diff --git a/adversarial/advbox/models/base.py b/adversarial/advbox/models/base.py
index 91b6fe4a3c97046839319b37708de4ef573ad655..2e5c397dc4706cc102e310202bca0a2254baa615 100644
--- a/adversarial/advbox/models/base.py
+++ b/adversarial/advbox/models/base.py
@@ -6,8 +6,8 @@ import abc
 
 abstractmethod = abc.abstractmethod
 
-class Model(object):
 
+class Model(object):
     """
     Base class of model to provide attack.
 
diff --git a/adversarial/advbox/models/paddle.py b/adversarial/advbox/models/paddle.py
index 831fa6a36278fea5d85104bc0a5c8ad4009af2e6..a72eb148bc691c41b804e6ed8b8a4430cc08efc5 100644
--- a/adversarial/advbox/models/paddle.py
+++ b/adversarial/advbox/models/paddle.py
@@ -7,6 +7,7 @@ from paddle.v2.fluid.framework import program_guard
 
 from .base import Model
 
+
 class PaddleModel(Model):
     """
     Create a PaddleModel instance.
@@ -30,9 +31,7 @@ class PaddleModel(Model):
                  channel_axis=3,
                  preprocess=None):
         super(PaddleModel, self).__init__(
-            bounds=bounds,
-            channel_axis=channel_axis,
-            preprocess=preprocess)
+            bounds=bounds, channel_axis=channel_axis, preprocess=preprocess)
 
         if preprocess is None:
             preprocess = (0, 1)
@@ -48,7 +47,8 @@ class PaddleModel(Model):
 
         # gradient
         loss = self._program.block(0).var(self._cost_name)
-        param_grads = fluid.backward.append_backward(loss, parameter_list=[self._input_name])
+        param_grads = fluid.backward.append_backward(
+            loss, parameter_list=[self._input_name])
         self._gradient = param_grads[0][1]
 
     def predict(self, image_batch):
@@ -61,16 +61,13 @@ class PaddleModel(Model):
                 numpy.ndarray: predictions of the images with shape (batch_size, num_of_classes).
         """
         feeder = fluid.DataFeeder(
-            feed_list=[self._input_name, self._logits_name], 
-            place=self._place, 
-            program=self._program
-        )
+            feed_list=[self._input_name, self._logits_name],
+            place=self._place,
+            program=self._program)
         predict_var = self._program.block(0).var(self._predict_name)
-        predict = self._exe.run(
-            self._program,
-            feed=feeder.feed(image_batch),
-            fetch_list=[predict_var]
-        )
+        predict = self._exe.run(self._program,
+                                feed=feeder.feed(image_batch),
+                                fetch_list=[predict_var])
         return predict
 
     def num_classes(self):
@@ -95,12 +92,10 @@ class PaddleModel(Model):
         """
         feeder = fluid.DataFeeder(
             feed_list=[self._input_name, self._logits_name],
-            place=self._place, 
-            program=self._program
-        )
-
-        grad, = self._exe.run(
-            self._program,
-            feed=feeder.feed(image_batch),
-            fetch_list=[self._gradient])
+            place=self._place,
+            program=self._program)
+
+        grad, = self._exe.run(self._program,
+                              feed=feeder.feed(image_batch),
+                              fetch_list=[self._gradient])
         return grad