diff --git a/README.md b/README.md
index a76829d..ac1d8e3 100644
--- a/README.md
+++ b/README.md
@@ -41,10 +41,16 @@ A face detection operator takes an image as input. it generates the bounding box
 
 ​	supported types: numpy.ndarray
 
-**Returns:**: *numpy.ndarray*
+**Returns:**: 
+
+*numpy.ndarray*
 
 ​	The detected face bounding boxes.
 
+*numpy.ndarray*
+
+​	The detected face bounding boxes confident scores.
+
 ## Code Example
 
 get detected face bounding boxes from './img1.jpg'. 
diff --git a/retinaface.py b/retinaface.py
index 58ab283..aab7280 100644
--- a/retinaface.py
+++ b/retinaface.py
@@ -22,10 +22,11 @@ from torchvision import transforms
 from pathlib import Path
 import numpy
 
-import towhee
+from towhee import register
 from towhee.operator import Operator
 from towhee.types.image_utils import to_pil
 from towhee._types import Image
+from towhee.types import arg, to_image_color
 
 from timm.data import resolve_data_config
 from timm.data.transforms_factory import create_transform
@@ -45,4 +46,5 @@ class Retinaface(Operator):
     def __call__(self, image: 'towhee._types.Image'):
         img = torch.FloatTensor(numpy.asarray(to_pil(image)))
         bboxes, keypoints = self.model(img)
-        return bboxes[:,:4],  bboxes[:,4]
+        bboxes = bboxes.cpu().detach().numpy()
+        return bboxes[:,:4].astype(int),  bboxes[:,4]