diff --git a/README.md b/README.md
index 0ec9ef4a6f6fc8a35918c8bc5bd3bc26dc59d53c..fd380d273dcee06ebf0765d355d89eab2749ef98 100644
--- a/README.md
+++ b/README.md
@@ -8,8 +8,9 @@ C. Farabet, C. Couprie, L. Najman, and Y. Lecun, “Learning hierarchical featur
 * scipy
 * sklearn
 * skimage
-* tensorflow (r1.0)
-* sacred
+* tensorflow==1.0
+* sacred==v0.7b2
+* pycoco (if using MS COCO database, https://github.com/pdollar/coco/tree/master/PythonAPI)
 
 If using GPU:
 * GPU enabled tensorflow
@@ -18,12 +19,12 @@ If using GPU:
 * cuDNN
 
 # How to use
-To run with synthetic squares dataset do:
-> python3 main.py
-To run with real scene data (Stanford backgournd dataset) do:
-> python3 main.py with dataset_name='stanford'
-To run using invariant-based kernels:
-> python3 main.py with invariants=True
+
+> `python3 experiment.py with mode=[test,train,evaluate,train_and_eval]`
+
+To see available params do:
+
+> `python3 experiment.py print_config`
 
 About the dataset:
 "The Stanford back- ground dataset [15] contains 715 images of outdoor scenes