add more description on original ResNet

Examples/Image/Miscellaneous/ImageNet/OriResNet/README.md

@@ -1,20 +1,24 @@
-# CNTK example: ImageNet ResNet 
+# CNTK example: Original ImageNet ResNet 
 
 ## Overview
+This work is an implementation of ResNet in CNTK. The work is strictly based on the original [ResNet paper](http://arxiv.org/abs/1512.03385). If you are interested in the original implementation of ResNet, follow [this link](https://github.com/KaimingHe/deep-residual-networks). 
 
+## Dataset
 |Data:     |The ILSVRC2012 dataset (http://www.image-net.org/challenges/LSVRC/2012/) of images.
 |:---------|:---
 |Purpose   |This example demonstrates usage of the NDL (Network Description Language) to define networks similar to ResNet.
 |Network   |NDLNetworkBuilder, deep convolutional residual networks (ResNet).
 |Training  |Stochastic gradient descent with momentum.
 
 ## Details
-The network configurations and experiment settings in this this folder resemble the ones in the original [ResNet paper](http://arxiv.org/abs/1512.03385) strictly without any extra optimization.
+* `Weight Decay in Batch Normalization`: Disable the weight decay in batch normalization. In our experiment, apply weight decay to all nodes will slow down the training curve convergence.
+* `Post Batch Normalization`: After training and before evaluating, using post batch normalization command to evaluate the mean and variance of batch normalization nodes instead of running statistics mean and variance. From the experiment, the statistics results of post batch normalization are more robust.
+
+## Results
 The following table contains results.
 
 | Network       | Top-1 error | Top-5 error | Model
 | ------------- | ----------- | ----------- | ----------
 | ResNet-50     | 24.58       | 7.43        |
 
-## Notes
-This work is an implementation of ResNets in CNTK. If you are interested in the original implementation of ResNet, follow [this link](https://github.com/KaimingHe/deep-residual-networks).
+## Notes