diff --git a/docs/featured_model/FACE_DETECTION.md b/docs/featured_model/FACE_DETECTION.md
index 4a141857c87f83cc9abe69f4f50820f07622e333..19db65d3f0469433e37ceda9a8ab3151fa363985 100644
--- a/docs/featured_model/FACE_DETECTION.md
+++ b/docs/featured_model/FACE_DETECTION.md
@@ -254,6 +254,12 @@ wget https://dataset.bj.bcebos.com/wider_face/wider_face_train_bbx_lmk_gt.txt
 
 (2)使用`configs/face_detection/blazeface_keypoint.yml`配置文件进行训练与评估，使用方法与上一节内容一致。
 
+### 模型评估
+
+| 网络结构     | 输入尺寸 | 图片个数/GPU | 学习率策略 | Easy Set  | Medium Set | Hard Set  | 下载 | 配置文件 |
+|:------------:|:----:|:-------:|:-------:|:---------:|:----------:|:---------:|:--------:|:--------:|
+| BlazeFace Keypoint     | 640  |    16    | 16w     | 0.852     | 0.816      | 0.662     | [模型](https://paddlemodels.bj.bcebos.com/object_detection/blazeface_keypoint.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/face_detection/blazeface_keypoint.yml) |
+
 ![](../images/12_Group_Group_12_Group_Group_12_84.jpg)
 
 ## 算法细节
@@ -285,7 +291,7 @@ wget https://dataset.bj.bcebos.com/wider_face/wider_face_train_bbx_lmk_gt.txt
 - 使用密度先验盒（density_prior_box）可提高检测精度。
 
 **版本信息:**
-- 原始版本: 参考原始论文复现；
+- 原始版本: 参考原始论文进行修改；
 - Lite版本: 使用更少的网络层数和通道数，具体可参考[代码](https://github.com/PaddlePaddle/PaddleDetection/blob/master/ppdet/modeling/architectures/faceboxes.py)。
 
 
diff --git a/docs/featured_model/FACE_DETECTION_en.md b/docs/featured_model/FACE_DETECTION_en.md
index 3bcc4a89e9e9a3f2d6d3a8c8fbc73114166970e5..a2cb4a2d4c44533641b094e06c26e17a7837a0fd 100644
--- a/docs/featured_model/FACE_DETECTION_en.md
+++ b/docs/featured_model/FACE_DETECTION_en.md
@@ -270,6 +270,12 @@ wget https://dataset.bj.bcebos.com/wider_face/wider_face_train_bbx_lmk_gt.txt
 
 (2)Use `configs/face_detection/blazeface_keypoint.yml` configuration file for training and evaluation, the method of use is the same as the previous section.
 
+### Evaluation
+
+| Architecture  | Size | Img/gpu | Lr schd | Easy Set  | Medium Set | Hard Set  | Download | Configs |
+|:------------:|:----:|:-------:|:-------:|:---------:|:----------:|:---------:|:--------:|:--------:|
+| BlazeFace Keypoint     | 640  |    16    | 16w     | 0.852     | 0.816      | 0.662     | [download](https://paddlemodels.bj.bcebos.com/object_detection/blazeface_keypoint.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/face_detection/blazeface_keypoint.yml) |
+
 ![](../images/12_Group_Group_12_Group_Group_12_84.jpg)
 
 
diff --git a/ppdet/modeling/architectures/blazeface.py b/ppdet/modeling/architectures/blazeface.py
index d740d07c09d6c5379a336a2360571982af5bf3a3..5090c90d6be6500cf5adbecd451de463c2b79409 100644
--- a/ppdet/modeling/architectures/blazeface.py
+++ b/ppdet/modeling/architectures/blazeface.py
@@ -41,6 +41,7 @@ class BlazeFace(object):
         output_decoder (object): `SSDOutputDecoder` instance
         min_sizes (list|None): min sizes of generated prior boxes.
         max_sizes (list|None): max sizes of generated prior boxes. Default: None.
+        steps (list|None): step size of adjacent prior boxes on each feature map.
         num_classes (int): number of output classes
         use_density_prior_box (bool): whether or not use density_prior_box
             instead of prior_box
diff --git a/ppdet/modeling/architectures/faceboxes.py b/ppdet/modeling/architectures/faceboxes.py
index b55354902238bb960e6ebf9605ec42c130e27b38..0c957501b23a8a55366b099db502fb44e4935b25 100644
--- a/ppdet/modeling/architectures/faceboxes.py
+++ b/ppdet/modeling/architectures/faceboxes.py
@@ -32,8 +32,8 @@ __all__ = ['FaceBoxes']
 @register
 class FaceBoxes(object):
     """
-    FaceBoxes: Sub-millisecond Neural Face Detection on Mobile GPUs,
-               see https://https://arxiv.org/abs/1708.05234
+    FaceBoxes: A CPU Real-time Face Detector with High Accuracy.
+               see https://arxiv.org/abs/1708.05234
 
     Args:
         backbone (object): backbone instance
@@ -42,7 +42,8 @@ class FaceBoxes(object):
             this attribute should be a list or tuple of integers.
         fixed_sizes (list|None): the fixed sizes of generated density prior boxes,
             this attribute should a list or tuple of same length with `densities`.
-        num_classes (int): number of output classes
+        num_classes (int): number of output classes.
+        steps (list|None): step size of adjacent prior boxes on each feature map.
     """
 
     __category__ = 'architecture'
@@ -55,7 +56,7 @@ class FaceBoxes(object):
                  densities=[[4, 2, 1], [1], [1]],
                  fixed_sizes=[[32., 64., 128.], [256.], [512.]],
                  num_classes=2,
-                 steps=[8., 16., 32.]):
+                 steps=[16., 32., 64.]):
         super(FaceBoxes, self).__init__()
         self.backbone = backbone
         self.num_classes = num_classes
@@ -116,7 +117,7 @@ class FaceBoxes(object):
                 fixed_ratios=[1.],
                 clip=False,
                 offset=0.5,
-                steps=[self.steps[i]] * 2)
+                steps=[self.steps[i]])
 
             num_boxes = box.shape[2]