diff --git a/model/model.py b/model/model.py
index 9265967..9b25aa3 100644
--- a/model/model.py
+++ b/model/model.py
@@ -329,6 +329,65 @@ def predict_step(self, session, data, summary=False, vis=False):
         return tag, ret_box3d_score
 
 
+    def save_frozen_graph(self, session, output_path):
+        """
+        saves a frozen graph.pb file in the given path
+        :param session: current tensorflow session
+        :param output_path: path to the output frozen_graph.pb
+        :return: ---
+        """
+        output_graph = tf.graph_util.convert_variables_to_constants(
+                session,
+                tf.get_default_graph().as_graph_def(),
+                self.get_output_nodes_names()
+                )
+
+        with tf.gfile.GFile(output_path, "wb") as f:
+            f.write(output_graph.SerializeToString())
+
+        print("%d ops in the final graph." % len(output_graph.node))
+        print("\n\n frozen graph saved to {}".format(output_path))
+
+    def get_output_nodes(self):
+        """
+        :return: list of all output nodes
+        """
+        ret = [
+                self.prob_output, self.delta_output,
+                self.box2d_ind_after_nms,
+                self.predict_summary
+        ]
+
+        return ret
+
+    def get_output_nodes_names(self):
+        """
+        :return: list of the names of all output nodes (string)
+        """
+        nodes = self.get_output_nodes()
+        ret = []
+        for nd in nodes:
+            ret.append(nd.name.split(":")[0])
+
+        print(ret)
+        return ret
+        
+
+    def get_input_nodes(self):
+        """
+        :return: list of all input nodes
+        """
+        ret = [
+                self.boxes2d, self.boxes2d_scores,
+                self.rgb, self.bv, self.bv_heatmap
+        ]
+        for idx in range(len(self.avail_gpus)):
+            ret.append(self.vox_feature[idx])
+            ret.append(self.vox_number[idx])
+            ret.append(self.vox_coordinate[idx])
+
+        return ret
+
 def average_gradients(tower_grads):
     # ref:
     # https://github.com/tensorflow/models/blob/6db9f0282e2ab12795628de6200670892a8ad6ba/tutorials/image/cifar10/cifar10_multi_gpu_train.py#L103
@@ -355,7 +414,8 @@ def average_gradients(tower_grads):
     return average_grads
 
 
+
+
 if __name__ == '__main__':
     pass
 
-