Correct a syntax error in README

Author: Thierry Dumas

README.md

@@ -55,11 +55,20 @@ cd autoencoder_based_image_compression/kodak_tensorflow/
    and H.265 are stored in the folder "eae/visualization/test/checking_reconstructing/kodak/".
 
 ## Quick start: training an autoencoder
-1. First of all, ImageNet images must be downloaded. In our case, it is sufficient to download the ILSVRC2012 validation images, "ILSVRC2012_img_val.tar" (6.3 GB), see [ImageNetDownloadWebPage](http://image-net.org/download). Let's say that, in your computer, the path to "ILSVRC2012_img_val.tar" is "path/to/folder_0/ILSVRC2012_img_val.tar" and you want the unpacked images to be put into the folder "path/to/folder_1/" before the script "creating_imagenet.py" preprocesses them. The creation of the ImageNet training and validaton sets of luminance images is then done via
+1. First of all, ImageNet images must be downloaded. In our case, it is sufficient to download the ILSVRC2012 validation
+   images, "ILSVRC2012_img_val.tar" (6.3 GB), see [ImageNetDownloadWebPage](http://image-net.org/download). Let's say that,
+   in your computer, the path to "ILSVRC2012_img_val.tar" is "path/to/folder_0/ILSVRC2012_img_val.tar" and you want the
+   unpacked images to be put into the folder "path/to/folder_1/" before the script "creating_imagenet.py" preprocesses
+   them. The creation of the ImageNet training and validaton sets of luminance images is then done via
    ```sh
    python creating_imagenet.py path/to/folder_1/ --path_to_tar=path/to/folder_0/ILSVRC2012_img_val.tar
    ```
-2. The training of an autoencoder on the ImageNet training set is done via the command below. 1.0 is the value of the quantization bin widths at the beginning of the training. 14000.0 is the value of the coefficient weighting the distortion term and the rate term in the objective function to be minimized over the parameters of the autoencoder. The script "training_eae_imagenet.py" enables to split the entire autoencoder training into several successive parts. The last argument 0 means that "training_eae_imagenet.py" runs the first part of the entire training. For each successive part, the last argument is incremented by 1.
+2. The training of an autoencoder on the ImageNet training set is done via the command below. 1.0 is the value of the
+   quantization bin widths at the beginning of the training. 14000.0 is the value of the coefficient weighting the
+   distortion term and the rate term in the objective function to be minimized over the parameters of the autoencoder.
+   The script "training_eae_imagenet.py" enables to split the entire autoencoder training into several successive parts.
+   The last argument 0 means that "training_eae_imagenet.py" runs the first part of the entire training. For each successive
+   part, the last argument is incremented by 1.
    ```sh
    python training_eae_imagenet.py 1.0 14000.0 0
    ```
@@ -68,7 +77,10 @@ cd autoencoder_based_image_compression/kodak_tensorflow/
 The documentation "documentation_kodak/documentation_code.html" describes all the functionalities of the code of the paper.
 
 ## A simple example
-Another piece of code is a simple example for introducing the code of the paper. This piece of code is stored in the folder "svhn". Its documentation is in the file "documentation_svhn/documentation_code.html". If you feel comfortable with autoencoders, this piece of code can be skipped. Its purpose is to clarify the training of a rate-distortion optimized autoencoder. That is why a simple rate-distortion optimized autoencoder with very few hidden units is trained on tiny images (32x32 SVHN digits).
+Another piece of code is a simple example for introducing the code of the paper. This piece of code is stored in the folder
+"svhn". Its documentation is in the file "documentation_svhn/documentation_code.html". If you feel comfortable with autoencoders,
+this piece of code can be skipped. Its purpose is to clarify the training of a rate-distortion optimized autoencoder. That is why
+a simple rate-distortion optimized autoencoder with very few hidden units is trained on tiny images (32x32 SVHN digits).
 
 ## Citing
 ```
@@ -78,3 +90,6 @@ Another piece of code is a simple example for introducing the code of the paper.
   booktitle = {ICASSP},
   year = {2018}
 }
+```
+
+