diff --git a/svo_analysis/README.md b/svo_analysis/README.md
index 0a4deeb..07ae6eb 100644
--- a/svo_analysis/README.md
+++ b/svo_analysis/README.md
@@ -25,3 +25,27 @@ Afterwards run
     rosrun svo_analysis benchmark.py flying_room_1_fast
 
 The script will create a new folder in `svo_analysis/results/`, run svo on the dataset and save the tracefile in this folder and at the end it will run some scripts to generate plots from the tracefile.
+
+
+#### Dataset file
+
+A dataset with motion-capture-system ground-truth should contain the following files
+
+* _images.txt_
+  
+     # timestamp img/imagename
+     1329330952.725747585296631 img/00000.png
+     1329330952.765715122222900 img/00001.png
+     1329330952.805604219436646 img/00002.png
+     ...
+
+* _groundtruth.txt_: groundtruth poses as acquired from a motion capture system
+
+     # timestamp tx ty tz qx qy qz qw
+     
+* _groundtruth_matched.txt_: the groundtruth pose associated to the images in images.txt. This file can be generated by the command:
+
+    cd my_dataset
+    rosrun svo_analysis associate.py images.txt groundtruth.txt
+
+