Update readme.md

Author: jiewen

readme.md

@@ -22,7 +22,7 @@
 ## The optimization process
 
 - The detailed information you can get from [there](https://github.com/ranjiewwen/TF_cifar10/blob/master/doc/scratch.md).
-- run scripts `tools/trian_cifar10.py` **include adjust lr , add data augmentation ,add dropout ,weight decay,stack 3*3 conv training tricks. you can learn how train model acc from 70%+ to 91+%**.
+- run scripts `tools/trian_cifar10.py` **include adjust lr , add data augmentation ,add dropout ,weight decay,stack 3*3 conv training tricks. you can learn how train model acc from 70%+ to 91+%**, while add model depth through `conv4_1 and conv4_2` it can not imporve val acc.
 ![](https://github.com/ranjiewwen/TF_cifar10/blob/master/doc/image/base_val_acc.png)
 - run scripts `tools/trian_cifar10_v2.py` include add batch_norm, we can see it make the training more unstable, maybe it not imporve val acc, while stack 3*3 conv it can improve val acc remarkable.
 ![](https://github.com/ranjiewwen/TF_cifar10/blob/master/doc/image/v2_val_acc.png)