diff --git a/docs/zh/week10/10-3.md b/docs/zh/week10/10-3.md
index f630f1178..b124c8c1f 100644
--- a/docs/zh/week10/10-3.md
+++ b/docs/zh/week10/10-3.md
@@ -219,7 +219,7 @@ emulator = nn.Sequential(
 
 <!--Refer to Figure 5. Block $\matr{C}$ represents the controller. It takes in the current state and ouputs a steering angle. Then block $\matr{T}$ (emulator) takes both the state and angle to produce the next state.-->
 
-根据图5，方块$\matr{C}$代表控制器。它输入进当前状态并输出驾驶角度。方块$\matr{T}$（模拟器）输送进状态和角度，产生下一状态。
+根据图6，方块$\matr{C}$代表控制器。它输入进当前状态并输出驾驶角度。方块$\matr{T}$（模拟器）输送进状态和角度，产生下一状态。
 
 |![]({{site.baseurl}}/images/week10/10-3/fig13.png)|
 | <center><b>Fig. 5</b>: 状态转换流图示</center>|
@@ -241,7 +241,7 @@ We then find the gradients using backpropagation and update parameters of the co
 ![]({{site.baseurl}}/images/week10/10-3/fig14.png)
 
 <!--To see this more clearly, we show the exact implementation of the emulator:-->
-圖6:「详细的模型结构」
+图7:「详细的模型结构」
 
 为了更好地理解，我们展示了准确的模拟器的执行过程：
 
@@ -271,7 +271,7 @@ criterion = nn.MSELoss()
 |![]({{site.baseurl}}/images/week10/10-3/fig16.png) | ![]({{site.baseurl}}/images/week10/10-3/fig17.png) |
 |![]({{site.baseurl}}/images/week10/10-3/fig18.png) | ![]({{site.baseurl}}/images/week10/10-3/fig19.png) |
 
-圖7:「运动示例」
+圖8:「运动示例」88
 
 ### <!--Additional Resources:-->补充资料：