Update with information about the convnets.

[pysvrt.git] / README.md

diff --git a/README.md b/README.md
index 26b40f2..9e350b8 100644 (file)
--- a/README.md
+++ b/README.md
@@ -6,7 +6,7 @@ framework.
 The main function is
 
 ```
 The main function is
 
 ```

-torch.ByteTensor generate_vignettes(int problem_number, torch.LongTensor labels)
+torch.ByteTensor svrt.generate_vignettes(int problem_number, torch.LongTensor labels)

 ```
 
 where
 ```
 
 where


@@ -32,5 +32,45 @@ make -j -k
 should generate an image example.png in the current directory.
 
 Note that the image generation does not take advantage of GPUs or
 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 3,000 vignettes per second and as
+multi-core, and can be as fast as 10,000 vignettes per second and as

 slow as 40 on a 4GHz i7-6700K.
 slow as 40 on a 4GHz i7-6700K.

+
+# Vignette compression #
+
+The two additional functions
+
+```
+torch.ByteStorage svrt.compress(torch.ByteStorage x)
+```
+
+and
+
+```
+torch.ByteStorage svrt.uncompress(torch.ByteStorage x)
+```
+
+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).
+
+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.
+
+# Testing convolution networks #
+
+The file
+
+```
+cnn-svrt.py
+```
+
+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.
+
+The networks were designed by Afroze Baqapuri during an internship at
+Idiap.