diff --git a/paddle/fluid/imperative/layer.h b/paddle/fluid/imperative/layer.h
index 1d109259f399a94323713f89e9c2333d7b720579..46107341a4edc3ec6bb5569c3fa50bbd9243bd26 100644
--- a/paddle/fluid/imperative/layer.h
+++ b/paddle/fluid/imperative/layer.h
@@ -141,11 +141,13 @@ class VarBase {
   void RunBackward();
 
   void TrackPreOp(OpBase* pre_op, const std::string& pre_op_out_name,
-                  int pre_op_out_idx, bool stop_gradient) {
+                  int pre_op_out_idx, bool pre_op_stop_gradient) {
     pre_op_ = pre_op;
     pre_op_out_name_ = pre_op_out_name;
     pre_op_out_idx_ = pre_op_out_idx;
-    stop_gradient_ = stop_gradient;
+    if (pre_op_stop_gradient) {
+      stop_gradient_ = pre_op_stop_gradient;
+    }
   }
 
   void ClearGradient() {
diff --git a/python/paddle/fluid/imperative/layers.py b/python/paddle/fluid/imperative/layers.py
index 57c45f764baa686fe749871ba948ad6bf581d2bc..c338c65a76b86cd28ee7cc8e38a422c4e382a9de 100644
--- a/python/paddle/fluid/imperative/layers.py
+++ b/python/paddle/fluid/imperative/layers.py
@@ -51,9 +51,8 @@ class Layer(core.Layer):
         return params
 
     def clear_gradients(self):
-        print([p.name for p in self.parameters()])
         for p in self.parameters():
-            if p.name not in set(['batch_norm_0.w_2', 'batch_norm_0.w_1']):
+            if not p._stop_gradient:
                 p._clear_gradient()
 
     def _build_once(self, inputs):
diff --git a/python/paddle/fluid/tests/unittests/test_imperative_resnet.py b/python/paddle/fluid/tests/unittests/test_imperative_resnet.py
index dfaaae0de3640dc3011a0f617896bfe790119861..c27fd0b8024a8fa3310a62de34299fb621e2902f 100644
--- a/python/paddle/fluid/tests/unittests/test_imperative_resnet.py
+++ b/python/paddle/fluid/tests/unittests/test_imperative_resnet.py
@@ -168,22 +168,22 @@ class ResNet(fluid.imperative.Layer):
         self.pool2d_max = Pool2D(
             pool_size=3, pool_stride=2, pool_padding=1, pool_type='max')
 
-        #  self.bottleneck_block_list = []
-        #  num_channels = 64
-        #  for block in range(len(depth)):
-        #  shortcut = False
-        #  for i in range(depth[block]):
-        #  bottleneck_block = BottleneckBlock(
-        #  num_channels=num_channels,
-        #  num_filters=num_filters[block],
-        #  stride=2 if i == 0 and block != 0 else 1,
-        #  shortcut=shortcut)
-        #  num_channels = bottleneck_block._num_channels_out
-        #  self.bottleneck_block_list.append(bottleneck_block)
-        #  shortcut = True
-
-        #  self.pool2d_avg = Pool2D(
-        #  pool_size=7, pool_type='avg', global_pooling=True)
+        self.bottleneck_block_list = []
+        num_channels = 64
+        for block in range(len(depth)):
+            shortcut = False
+            for i in range(depth[block]):
+                bottleneck_block = BottleneckBlock(
+                    num_channels=num_channels,
+                    num_filters=num_filters[block],
+                    stride=2 if i == 0 and block != 0 else 1,
+                    shortcut=shortcut)
+                num_channels = bottleneck_block._num_channels_out
+                self.bottleneck_block_list.append(bottleneck_block)
+                shortcut = True
+
+        self.pool2d_avg = Pool2D(
+            pool_size=7, pool_type='avg', global_pooling=True)
 
         import math
         stdv = 1.0 / math.sqrt(2048 * 1.0)
@@ -196,9 +196,9 @@ class ResNet(fluid.imperative.Layer):
     def forward(self, inputs):
         y = self.conv(inputs)
         y = self.pool2d_max(y)
-        #  for bottleneck_block in self.bottleneck_block_list:
-        #  y = bottleneck_block(y)
-        #  y = self.pool2d_avg(y)
+        for bottleneck_block in self.bottleneck_block_list:
+            y = bottleneck_block(y)
+        y = self.pool2d_avg(y)
         y = self.out(y)
         return y
 
@@ -209,7 +209,7 @@ class TestImperativeResnet(unittest.TestCase):
 
         batch_size = train_parameters["batch_size"]
         batch_num = 1
-        with fluid.imperative.guard(place=fluid.CPUPlace()):
+        with fluid.imperative.guard():
             fluid.default_startup_program().random_seed = seed
             fluid.default_main_program().random_seed = seed
 
@@ -275,9 +275,8 @@ class TestImperativeResnet(unittest.TestCase):
             fluid.default_startup_program().random_seed = seed
             fluid.default_main_program().random_seed = seed
 
-            exe = fluid.Executor(fluid.CPUPlace())
-            #  exe = fluid.Executor(fluid.CPUPlace(
-            #  ) if not core.is_compiled_with_cuda() else fluid.CUDAPlace(0))
+            exe = fluid.Executor(fluid.CPUPlace(
+            ) if not core.is_compiled_with_cuda() else fluid.CUDAPlace(0))
 
             resnet = ResNet()
             optimizer = optimizer_setting(train_parameters)
@@ -347,7 +346,6 @@ class TestImperativeResnet(unittest.TestCase):
                     static_grad_value[static_grad_name_list[
                         i - grad_start_pos]] = out[i]
 
-        print(static_out, dy_out)
         self.assertTrue(np.allclose(static_out, dy_out))
 
         self.assertEqual(len(dy_param_init_value), len(static_param_init_value))
@@ -358,9 +356,7 @@ class TestImperativeResnet(unittest.TestCase):
 
         self.assertEqual(len(dy_grad_value), len(static_grad_value))
         for key, value in six.iteritems(static_grad_value):
-            if not np.allclose(value, dy_grad_value[key]):
-                print(key)
-            #self.assertTrue(np.allclose(value, dy_grad_value[key]))
+            self.assertTrue(np.allclose(value, dy_grad_value[key]))
             self.assertTrue(np.isfinite(value.all()))
             self.assertFalse(np.isnan(value.any()))