diff --git a/mace/python/tools/converter_tool/tensorflow_converter.py b/mace/python/tools/converter_tool/tensorflow_converter.py
index ac3333c7d597f4e62e98a934eb16fdf6b61c0f07..a4430841d5de678f2bbf37a295622e04c3c4a8c8 100644
--- a/mace/python/tools/converter_tool/tensorflow_converter.py
+++ b/mace/python/tools/converter_tool/tensorflow_converter.py
@@ -200,6 +200,8 @@ class TensorflowConverter(base_converter.ConverterInterface):
         tf_graph_def = tf.GraphDef()
         with tf.gfile.Open(src_model_file, 'rb') as f:
             tf_graph_def.ParseFromString(f.read())
+
+        self._placeholders = {}
         self.add_shape_info(tf_graph_def)
 
         with tf.Session() as session:
@@ -240,6 +242,8 @@ class TensorflowConverter(base_converter.ConverterInterface):
                         tensor_shape_pb2.TensorShapeProto.Dim(size=i) for i in
                         input_node.shape
                     ])
+                    self._placeholders[node.name + ':0'] = \
+                        np.zeros(shape=input_node.shape, dtype=float)
 
     @staticmethod
     def get_scope(tensor_name):
@@ -288,13 +292,18 @@ class TensorflowConverter(base_converter.ConverterInterface):
 
     # this function tries to infer tensor shape, but some dimension shape
     # may be undefined due to variance of input length
-    @staticmethod
-    def infer_tensor_shape(tensor):
-        shape = tensor.shape.as_list()
-
-        def normalize_func(dim):
-            return dim if dim else - 1
-        return [normalize_func(dim) for dim in shape]
+    def infer_tensor_shape(self, tensor):
+        inferred_tensor_shape = tensor.shape.as_list()
+        inferred_success = True
+        for _, dim in enumerate(inferred_tensor_shape):
+            if dim is None:
+                inferred_success = False
+                break
+        if inferred_success:
+            return inferred_tensor_shape
+
+        tensor_shape = tf.shape(tensor).eval(feed_dict=self._placeholders)
+        return tensor_shape
 
     def convert_nop(self, tf_op):
         pass