X-Git-Url: https://www.fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=README.md;h=4efe67fbc4538875228fdacddbb3ab38c3c9e113;hb=f2da660a51ed51cab4d81df29109e26e0da347bc;hp=735bee39a9514619a4f4e158ad4fa0927149d17c;hpb=05414734a8c423314aaa9447db04ed348066c2f9;p=pysvrt.git

diff --git a/README.md b/README.md
index 735bee3..4efe67f 100644
--- a/README.md
+++ b/README.md
@@ -1,7 +1,8 @@
 # Introduction #
 
-This is the port of the Synthetic Visual Reasoning Test to the pytorch
-framework.
+This is a port of the Synthetic Visual Reasoning Test problems to the
+pytorch framework, with an implementation of two convolutional
+networks to solve them.
 
 The main function is
 
@@ -20,6 +21,23 @@ The returned ByteTensor has three dimensions:
  * Pixel row
  * Pixel col
 
+# Installation and test #
+
+Executing
+
+```
+make -j -k
+./test-svrt.py
+```
+
+should generate an image example.png in the current directory.
+
+Note that the image generation does not take advantage of GPUs or
+multi-core, and can be as fast as 10,000 vignettes per second and as
+slow as 40 on a 4GHz i7-6700K.
+
+# Vignette compression #
+
 The two additional functions
 
 ```
@@ -36,23 +54,24 @@ provide a lossless compression scheme adapted to the ByteStorage of
 the vignette ByteTensor (i.e. expecting a lot of 255s, a few 0s, and
 no other value).
 
-They allow to reduce the memory footprint by a factor ~50, and may be
-usefull to deal with very large data-sets and avoid re-generating
-images at every batch.
+This compression reduces the memory footprint by a factor ~50, and may
+be usefull to deal with very large data-sets and avoid re-generating
+images at every batch. It induces a little overhead for decompression,
+and moving from CPU to GPU memory.
 
 See vignette_set.py for a class CompressedVignetteSet using it.
 
-# Installation and test #
+# Testing convolution networks #
 
-Executing
+The file
 
 ```
-make -j -k
-./test-svrt.py
+cnn-svrt.py
 ```
 
-should generate an image example.png in the current directory.
+provides the implementation of two deep networks, and use the
+compressed vignette code to allow the training with several millions
+vignettes on a PC with 16Gb and a GPU with 8Gb.
 
-Note that the image generation does not take advantage of GPUs or
-multi-core, and can be as fast as 10,000 vignettes per second and as
-slow as 40 on a 4GHz i7-6700K.
+The networks were designed by Afroze Baqapuri during an internship at
+Idiap.