diff --git a/src/TensorFlowNET.Core/Gradients/nn_grad.cs b/src/TensorFlowNET.Core/Gradients/nn_grad.cs
index 967b3c21..e4502ad8 100644
--- a/src/TensorFlowNET.Core/Gradients/nn_grad.cs
+++ b/src/TensorFlowNET.Core/Gradients/nn_grad.cs
@@ -170,6 +170,14 @@ namespace Tensorflow.Gradients
         public static Tensor[] _FusedBatchNormGrad(Operation op, Tensor[] grads)
             => _BaseFusedBatchNormGrad(op, 0, grads);
 
+        [RegisterGradient("FusedBatchNormV2")]
+        public static Tensor[] _FusedBatchNormV2Grad(Operation op, Tensor[] grads)
+            => _BaseFusedBatchNormGrad(op, 1, grads);
+
+        [RegisterGradient("FusedBatchNormV3")]
+        public static Tensor[] _FusedBatchNormV3Grad(Operation op, Tensor[] grads)
+            => _BaseFusedBatchNormGrad(op, 2, grads);
+
         /// <summary>
         /// Return the gradients for the 3 inputs of BatchNorm.
         /// </summary>
@@ -190,8 +198,10 @@ namespace Tensorflow.Gradients
             switch (version)
             {
                 case 2:
-                    throw new NotImplementedException("");
+                    grad_fun = gen_nn_ops.fused_batch_norm_grad_v3;
+                    break;
                 case 1:
+                    // grad_fun = gen_nn_ops.fused_batch_norm_grad_v2;
                     throw new NotImplementedException("");
                 default:
                     grad_fun = gen_nn_ops.fused_batch_norm_grad;
@@ -225,8 +235,8 @@ namespace Tensorflow.Gradients
                     YBackprop = grad_y,
                     X = x,
                     Scale = scale,
-                    ReserveSpace1 = op.outputs[3],
-                    ReserveSpace2 = op.outputs[4],
+                    ReserveSpace1 = pop_mean,
+                    ReserveSpace2 = pop_var,
                     ReserveSpace3 = version == 2 ? op.outputs[5] : null,
                     Epsilon = epsilon,
                     DataFormat = data_format,
diff --git a/src/TensorFlowNET.Core/Operations/OpDefLibrary.cs b/src/TensorFlowNET.Core/Operations/OpDefLibrary.cs
index 5700ccdd..e842fcb4 100644
--- a/src/TensorFlowNET.Core/Operations/OpDefLibrary.cs
+++ b/src/TensorFlowNET.Core/Operations/OpDefLibrary.cs
@@ -65,6 +65,10 @@ namespace Tensorflow
                 var base_types = new List<TF_DataType>();
                 var types = new List<TF_DataType>();
 
+#if DEBUG
+                if (op_type_name == "FusedBatchNormGradV3")
+                    ;
+#endif
                 // Perform input type inference
                 foreach (var input_arg in op_def.InputArg)
                 {
diff --git a/src/TensorFlowNET.Core/Operations/Operation.Implicit.cs b/src/TensorFlowNET.Core/Operations/Operation.Implicit.cs
index 8de412c8..9cadac0c 100644
--- a/src/TensorFlowNET.Core/Operations/Operation.Implicit.cs
+++ b/src/TensorFlowNET.Core/Operations/Operation.Implicit.cs
@@ -32,7 +32,7 @@ namespace Tensorflow
 
         public override string ToString()
         {
-            return _handle == IntPtr.Zero ? "tf.Operation Undefined" : $"tf.Operation '{name}' type={OpType}";
+            return _handle == IntPtr.Zero ? "tf.Operation Undefined" : $"<tf.Operation '{name}' type={OpType}>";
         }
 
         public override bool Equals(object obj)
diff --git a/src/TensorFlowNET.Core/Operations/Operation.Input.cs b/src/TensorFlowNET.Core/Operations/Operation.Input.cs
index fdf92504..bbd11b0e 100644
--- a/src/TensorFlowNET.Core/Operations/Operation.Input.cs
+++ b/src/TensorFlowNET.Core/Operations/Operation.Input.cs
@@ -80,6 +80,9 @@ namespace Tensorflow
         /// reasons, or to ensure that the side effects of an op are observed
         /// in the correct order.
         /// </summary>
+#if SERIALIZABLE
+        [JsonIgnore]
+#endif
         public Operation[] control_inputs
         {
             get
diff --git a/src/TensorFlowNET.Core/Operations/Operation.Output.cs b/src/TensorFlowNET.Core/Operations/Operation.Output.cs
index abe8e9c1..3a70c107 100644
--- a/src/TensorFlowNET.Core/Operations/Operation.Output.cs
+++ b/src/TensorFlowNET.Core/Operations/Operation.Output.cs
@@ -45,6 +45,9 @@ namespace Tensorflow
         }
 
         private Tensor[] _outputs;
+#if SERIALIZABLE
+        [JsonIgnore]
+#endif
         public Tensor[] outputs => _outputs;
 #if SERIALIZABLE
         [JsonIgnore]
diff --git a/src/TensorFlowNET.Core/Operations/Operation.cs b/src/TensorFlowNET.Core/Operations/Operation.cs
index 831e6ca5..a0acf4bb 100644
--- a/src/TensorFlowNET.Core/Operations/Operation.cs
+++ b/src/TensorFlowNET.Core/Operations/Operation.cs
@@ -74,6 +74,9 @@ namespace Tensorflow
         public TF_DataType dtype => TF_DataType.DtInvalid;
         public string name => _handle == IntPtr.Zero ? null : c_api.StringPiece(c_api.TF_OperationName(_handle));
         public string OpType => _handle == IntPtr.Zero ? null : c_api.StringPiece(c_api.TF_OperationOpType(_handle));
+#if SERIALIZABLE
+        [JsonIgnore]
+#endif
         public string Device => _handle == IntPtr.Zero ? null : c_api.StringPiece(c_api.TF_OperationDevice(_handle));
 #if SERIALIZABLE
         [JsonIgnore]
@@ -152,6 +155,10 @@ namespace Tensorflow
         {
             _graph = g;
 
+#if DEBUG
+            if (node_def.Name == "define_second_stage_train/gradients/define_loss/conv_lobj_branch/batch_normalization/cond/FusedBatchNormV3_1_grad/FusedBatchNormGradV3")
+                ;
+#endif
             // Build the list of control inputs.
             var control_input_ops = new List<Operation>();
             if (control_inputs != null)
diff --git a/src/TensorFlowNET.Core/Operations/array_ops.cs b/src/TensorFlowNET.Core/Operations/array_ops.cs
index f9f2f58f..5374d72b 100644
--- a/src/TensorFlowNET.Core/Operations/array_ops.cs
+++ b/src/TensorFlowNET.Core/Operations/array_ops.cs
@@ -427,7 +427,7 @@ namespace Tensorflow
                 if (!tf.context.executing_eagerly())
                 {
                     var input_tensor = ops.convert_to_tensor(input);
-                    var input_shape = tensor_util.to_shape(input_tensor.shape);
+                    var input_shape = input_tensor.TensorShape;
                     if (optimize && input_tensor.NDims > -1 && input_shape.is_fully_defined())
                     {
                         var nd = np.array(input_tensor.shape).astype(out_type.as_numpy_dtype());
diff --git a/src/TensorFlowNET.Core/Operations/gen_math_ops.cs b/src/TensorFlowNET.Core/Operations/gen_math_ops.cs
index 5cf240e8..7621d1b2 100644
--- a/src/TensorFlowNET.Core/Operations/gen_math_ops.cs
+++ b/src/TensorFlowNET.Core/Operations/gen_math_ops.cs
@@ -141,7 +141,8 @@ namespace Tensorflow
 
         public static Tensor add<Tx, Ty>(Tx x, Ty y, string name = null)
         {
-            var _op = _op_def_lib._apply_op_helper("Add", name, args: new { x, y });
+            // forward_compatible(2019, 6, 25):
+            var _op = _op_def_lib._apply_op_helper("AddV2", name, args: new { x, y });
 
             return _op.output;
         }
diff --git a/src/TensorFlowNET.Core/Operations/math_ops.cs b/src/TensorFlowNET.Core/Operations/math_ops.cs
index bb8d7134..bc904028 100644
--- a/src/TensorFlowNET.Core/Operations/math_ops.cs
+++ b/src/TensorFlowNET.Core/Operations/math_ops.cs
@@ -355,6 +355,9 @@ namespace Tensorflow
             return _may_reduce_to_scalar(keepdims, axis, all);
         }
 
+        public static Tensor realdiv(Tensor x, Tensor y, string name = null)
+            => gen_math_ops.real_div(x, y, name: name);
+
         /// <summary>
         /// Computes log(sum(exp(elements across dimensions of a tensor))).
         /// Reduces `input_tensor` along the dimensions given in `axis`.
@@ -561,6 +564,9 @@ namespace Tensorflow
         public static Tensor rsqrt(Tensor x, string name = null)
             => gen_math_ops.rsqrt(x, name: name);
 
+        public static Tensor pow<Tx, Ty>(Tx x, Ty y, string name = null)
+            => gen_math_ops.pow(x, y, name: name);
+
         public static Tensor range(object start, object limit = null, object delta = null, TF_DataType dtype = TF_DataType.DtInvalid, string name = "range")
         {
             if(limit == null)
diff --git a/src/TensorFlowNET.Core/Operations/nn_impl.py.cs b/src/TensorFlowNET.Core/Operations/nn_impl.py.cs
index 42103b00..a6c9e221 100644
--- a/src/TensorFlowNET.Core/Operations/nn_impl.py.cs
+++ b/src/TensorFlowNET.Core/Operations/nn_impl.py.cs
@@ -117,7 +117,7 @@ namespace Tensorflow
             var min_epsilon = 1.001e-5f;
             epsilon = epsilon > min_epsilon ? epsilon : min_epsilon;
 
-            var results = gen_nn_ops.fused_batch_norm(x,
+            var results = gen_nn_ops.fused_batch_norm_v3(x,
                 scale_tensor,
                 offset_tensor,
                 mean,
diff --git a/src/TensorFlowNET.Core/TensorFlow.Binding.csproj b/src/TensorFlowNET.Core/TensorFlow.Binding.csproj
index 09e2674a..369c6c81 100644
--- a/src/TensorFlowNET.Core/TensorFlow.Binding.csproj
+++ b/src/TensorFlowNET.Core/TensorFlow.Binding.csproj
@@ -33,7 +33,7 @@ https://tensorflownet.readthedocs.io</Description>
 
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|AnyCPU'">
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <DefineConstants>TRACE;DEBUG;SERIALIZABLE_</DefineConstants>
+    <DefineConstants>TRACE;DEBUG;SERIALIZABLE</DefineConstants>
   </PropertyGroup>
 
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|AnyCPU'">
@@ -62,6 +62,7 @@ https://tensorflownet.readthedocs.io</Description>
 
   <ItemGroup>
     <PackageReference Include="Google.Protobuf" Version="3.11.3" />
+    <PackageReference Include="Newtonsoft.Json" Version="12.0.3" />
     <PackageReference Include="NumSharp.Lite" Version="0.1.0" />
     <PackageReference Include="Protobuf.Text" Version="0.4.0" />
   </ItemGroup>
diff --git a/src/TensorFlowNET.Core/Tensors/Tensor.cs b/src/TensorFlowNET.Core/Tensors/Tensor.cs
index efac802d..2ec02232 100644
--- a/src/TensorFlowNET.Core/Tensors/Tensor.cs
+++ b/src/TensorFlowNET.Core/Tensors/Tensor.cs
@@ -115,6 +115,9 @@ namespace Tensorflow
         /// <summary>
         ///     The name of the device on which this tensor will be produced, or null.
         /// </summary>
+#if SERIALIZABLE
+        [JsonIgnore]
+#endif
         public string Device => op.Device;
 #if SERIALIZABLE
         [JsonIgnore]
diff --git a/src/TensorFlowNET.Core/ops.name_scope.cs b/src/TensorFlowNET.Core/ops.name_scope.cs
index 80397667..55e9cf61 100644
--- a/src/TensorFlowNET.Core/ops.name_scope.cs
+++ b/src/TensorFlowNET.Core/ops.name_scope.cs
@@ -68,7 +68,7 @@ namespace Tensorflow
                 var g = get_default_graph();
                 g._name_stack = old_stack;
             }
-
+            
             public void __exit__()
             {
             }
@@ -82,7 +82,7 @@ namespace Tensorflow
             {
                 
             }
-
+            
             /// <summary>
             /// __enter__()
             /// </summary>
diff --git a/test/TensorFlowNET.UnitTest/Hub/MnistModelLoaderTest.cs b/test/TensorFlowNET.UnitTest/Hub/MnistModelLoaderTest.cs
index 26dfd3b6..b1c90b32 100644
--- a/test/TensorFlowNET.UnitTest/Hub/MnistModelLoaderTest.cs
+++ b/test/TensorFlowNET.UnitTest/Hub/MnistModelLoaderTest.cs
@@ -2,7 +2,7 @@ using Microsoft.VisualStudio.TestTools.UnitTesting;
 using System.Threading.Tasks;
 using Tensorflow.Hub;
 
-namespace UnitTest
+namespace TensorFlowNET.UnitTest
 {
     [TestClass]
     public class MnistModelLoaderTest