diff --git a/configs/anchor_free/README.md b/configs/anchor_free/README.md
index 7c0d196c60324d64864fdedfa605f101ac21e253..184ca2a6e938cc37d3a57e684480e7c0d035beac 100644
--- a/configs/anchor_free/README.md
+++ b/configs/anchor_free/README.md
@@ -38,6 +38,7 @@
 - CornerNet-Squeeze中使用ResNet结构的骨干网络时，加入了FPN结构，骨干网络的输出feature map采用FPN中的P3层输出。
 - \*CornerNet-Squeeze-dcn-mixup-cosine是基于原版CornerNet-Squeeze优化效果最好的模型，在ResNet的骨干网络基础上增加mixup预处理和使用cosine_decay
 - FCOS使用GIoU loss、用location分支预测centerness、左上右下角点偏移量归一化和ground truth中心匹配策略
+- Cornernet-Squeeze模型依赖corner_pooling op，该op在```ppdet/ext_op```中编译得到，具体编译方式请参考[自定义OP的编译过程](../../ppdet/ext_op/README.md)
 
 ## 算法细节
 
diff --git a/ppdet/ext_op/README.md b/ppdet/ext_op/README.md
index cc6173b36c3d70d4e30706da12b32523dcc6e00c..6bfeb0e8843abeacedc7e062c0eb5636124a4644 100644
--- a/ppdet/ext_op/README.md
+++ b/ppdet/ext_op/README.md
@@ -37,6 +37,23 @@ sh make.sh
 编译自定义OP时会报错找不到`mkldnn.h`等文件，可在`make.sh`中删除编译命令中的`-DPADDLE_WITH_MKLDNN`选项。
 
 
+## 设置环境变量
+
+需要将Paddle的核心库设置到`LD_LIBRARY_PATH`里, 先运行下面程序获取路径:
+
+```
+import paddle
+print(paddle.sysconfig.get_lib())
+```
+
+可通过如下方式添加动态库路径:
+
+```
+export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:`python -c 'import paddle; print(paddle.sysconfig.get_lib())'`
+```
+
+
+
 ## 执行单测
 
 执行下列单测，确保自定义算子可在网络中正确使用：
diff --git a/ppdet/ext_op/src/make.sh b/ppdet/ext_op/src/make.sh
index bd0d3a3c904bd9409d200100de6d8e30ca974504..ffc66034f89587806044972335b1a40354407264 100755
--- a/ppdet/ext_op/src/make.sh
+++ b/ppdet/ext_op/src/make.sh
@@ -19,5 +19,3 @@ g++ bottom_pool_op.cc bottom_pool_op.cu.o top_pool_op.cc top_pool_op.cu.o right_
   -L /usr/local/cuda/lib64 -lpaddle_framework -lcudart
 
 rm *.cu.o
-
-export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$lib_dir
diff --git a/ppdet/modeling/anchor_heads/corner_head.py b/ppdet/modeling/anchor_heads/corner_head.py
index a43416d15c3b110593ae51e8d77ad5866725959c..01e1195b0c68bb66837db29a69ab6e6b9f9f5b6e 100644
--- a/ppdet/modeling/anchor_heads/corner_head.py
+++ b/ppdet/modeling/anchor_heads/corner_head.py
@@ -246,7 +246,8 @@ class CornerHead(object):
         try:
             import cornerpool_lib
         except:
-            logger.error("cornerpool_lib not found, compile in ext_op at first")
+            logger.error(
+                "cornerpool_lib not found, compile in ppdet/ext_op at first")
         self.train_batch_size = train_batch_size
         self.test_batch_size = test_batch_size
         self.num_classes = num_classes