diff --git a/detic/predictor.py b/detic/predictor.py
index 318205a..7d116c7 100644
--- a/detic/predictor.py
+++ b/detic/predictor.py
@@ -5,6 +5,7 @@
 from collections import deque
 import cv2
 import torch
+import random
 
 from detectron2.data import MetadataCatalog
 from detectron2.engine.defaults import DefaultPredictor
@@ -37,7 +38,7 @@ def get_clip_embeddings(vocabulary, prompt='a '):
 }
 
 class VisualizationDemo(object):
-    def __init__(self, cfg, args, 
+    def __init__(self, cfg, args,
         instance_mode=ColorMode.IMAGE, parallel=False):
         """
         Args:
@@ -50,10 +51,12 @@ def __init__(self, cfg, args,
             self.metadata = MetadataCatalog.get("__unused")
             self.metadata.thing_classes = args.custom_vocabulary.split(',')
             classifier = get_clip_embeddings(self.metadata.thing_classes)
+            self._default_vocabulary = None
         else:
             self.metadata = MetadataCatalog.get(
                 BUILDIN_METADATA_PATH[args.vocabulary])
             classifier = BUILDIN_CLASSIFIER[args.vocabulary]
+            self._default_vocabulary = args.vocabulary
 
         num_classes = len(self.metadata.thing_classes)
         self.cpu_device = torch.device("cpu")
@@ -67,6 +70,26 @@ def __init__(self, cfg, args,
             self.predictor = DefaultPredictor(cfg)
         reset_cls_test(self.predictor.model, classifier, num_classes)
 
+    def change_vocabulary(self, vocab):
+        """
+        Args:
+            vocab (str): The comma separated string of vocabulary
+        """
+        self.metadata = MetadataCatalog.get("__unused+"+str(random.random()))
+        self.metadata.thing_classes = vocab.split(',')
+        classifier = get_clip_embeddings(self.metadata.thing_classes)
+        num_classes = len(self.metadata.thing_classes)
+        reset_cls_test(self.predictor.model, classifier, num_classes)
+
+    def set_defalt_vocabulary(self):
+        if not self._default_vocabulary:
+            raise RuntimeError("The VisualizationDemo is not initalized with buildin vocabulary")
+        self.metadata = MetadataCatalog.get(
+                BUILDIN_METADATA_PATH[self._default_vocabulary])
+        classifier = BUILDIN_CLASSIFIER[self._default_vocabulary]
+        num_classes = len(self.metadata.thing_classes)
+        reset_cls_test(self.predictor.model, classifier, num_classes)
+
     def run_on_image(self, image):
         """
         Args: