diff --git a/README.md b/README.md
index a95d679..1d654bb 100644
--- a/README.md
+++ b/README.md
@@ -38,7 +38,7 @@ BM_bireal18_imagenet        61809506 ns   61056865 ns         10       <--- Bi-r
 BM_bireal18_imagenet_stem   43279353 ns   41533009 ns         14       <--- Bi-real Net 18 with stem module (The network structure will be described in detail in the coming paper), 56.4% top-1 on ImageNet
 ```
 
-The following is the comparison between our dabnn and [Caffe](http://caffe.berkeleyvision.org) (full precision), [TensorFlow Lite](https://www.tensorflow.org/lite) (full precision) and [BMXNet](https://github.com/hpi-xnor/BMXNet) (binary). We surprisingly observe that BMXNet is even slower the full precision TensorFlow Lite. It suggests that the potential of binary neural networks is far from exploited until our dabnn is published.
+The following is the comparison between our dabnn and [Caffe](http://caffe.berkeleyvision.org) (full precision), [TensorFlow Lite](https://www.tensorflow.org/lite) (full precision) and [BMXNet](https://github.com/hpi-xnor/BMXNet) (binary). Note that "Conv 64", "Conv 128", "Conv 256" and "Conv 512" have the same meaning as in the above benchmark. We surprisingly observe that BMXNet is even slower the full precision TensorFlow Lite. It suggests that the potential of binary neural networks is far from exploited until our dabnn is published.
 
 ![Comparison](images/comparison_en.png)
 
diff --git a/README_CN.md b/README_CN.md
index b79122d..e75eeeb 100644
--- a/README_CN.md
+++ b/README_CN.md
@@ -38,7 +38,7 @@ BM_bireal18_imagenet        61809506 ns   61056865 ns         10       <--- Bi-r
 BM_bireal18_imagenet_stem   43279353 ns   41533009 ns         14       <--- 带有 stem 模块的 Bi-real Net 18 (将在论文中描述), ImageNet top-1 为 56.4%
 ```
 
-在 Google Pixel 1 上与 [Caffe](http://caffe.berkeleyvision.org)（全精度）, [TensorFlow Lite](https://www.tensorflow.org/lite)（全精度）和 [BMXNet](https://github.com/hpi-xnor/BMXNet)（二值）的对比如下。我们很惊讶的发现现有的二值 inference 框架 BMXNet 甚至比全精度的 TensorFlow Lite 还要慢，这表明，直到 dabnn 推出之前，二值网络的潜力都远远没有被挖掘出来。
+在 Google Pixel 1 上与 [Caffe](http://caffe.berkeleyvision.org)（全精度）, [TensorFlow Lite](https://www.tensorflow.org/lite)（全精度）和 [BMXNet](https://github.com/hpi-xnor/BMXNet)（二值）的对比如下，其中 Conv 64、Conv 128、Conv 256 和 Conv 512 和上面的 benchmark 中的含义相同。我们很惊讶的发现现有的二值 inference 框架 BMXNet 甚至比全精度的 TensorFlow Lite 还要慢，这表明，直到 dabnn 推出之前，二值网络的潜力都远远没有被挖掘出来。
 
 ![Comparison](images/comparison_cn.png)