Update README.md

README.md

@@ -5,8 +5,18 @@ This repo contains the experiment code in paper
 
 The paper presents a new zero-shot learning method, which achieves the state-of-the-art results in ImageNet fall2011.
 
-In order to reproduce the experiments, you need to download the following files from http://image-net.org/
+In order to reproduce the experiments, first you need to download the following files from http://image-net.org/
 
 - ILSVRC2012_img_train.tar (138G)
 - ILSVRC2012_img_val.tar (6.3G)
 - fall11_whole.tar (1.2T)
+
+Second, you need to prepare the images intro HDF5 files, using 
+- uncompress.sh
+- correct_format.sh
+- image2hdf5.sh
+
+Third, to compute the CNN outputs of GoogLeNet of the images, using
+- caffe_outputs.py
+
+