diff --git a/README.md b/README.md
index 59c08a5..7206700 100644
--- a/README.md
+++ b/README.md
@@ -1,9 +1,17 @@
 ## High-Resolution Image Inpainting using Multi-Scale Neural Patch Synthesis
 
+
 ![teaser](images/teaser.png "Sample inpainting results on held-out ImageNet images")
 
 This is the code for [High-Resolution Image Inpainting using Multi-Scale Neural Patch Synthesis](https://arxiv.org/pdf/1611.09969). Given an image, we use the content and texture network to jointly infer the missing region. This repository contains the pre-trained model for the content network and the joint optimization code, including the demo to run example images. The code is adapted from the [Context Encoders](https://github.com/pathak22/context-encoder) and [CNNMRF](https://github.com/chuanli11/CNNMRF). Please contact [Harry Yang](http://www.harryyang.org) for questions regarding the paper or the code. Note that the code is for research purpose only.
 
+**[update 9/16/2017]** We increase the speed of original version by 6x (30s/image on GPU). The updates include:
+
+1. We remove layer 12 of vgg in texture optimization.
+2. In texture optimization, we have three scales and 100 iterations at each scale. Now it only computes the nearest patch at the first iteration of each scale, and re-use the nearest index in later iterations.
+
+This greatly increases the speed at the cost of very subtle inpainting quality. 
+
 ### Demo
 
 - Install Torch:  http://torch.ch/docs/getting-started.html#_