diff --git a/README.md b/README.md
index 51a9b0b..9e49320 100644
--- a/README.md
+++ b/README.md
@@ -11,6 +11,12 @@ Existing vision-based action recognition is susceptible to occlusion and appeara
 ## Datasets
 [UTD-MHAD](https://personal.utdallas.edu/~kehtar/UTD-MHAD.html)        
 [Berkeley-MHAD](https://tele-immersion.citris-uc.org/berkeley_mhad/)           
+[MMAct](https://mmact19.github.io/2019/)       
+
+## GAF Data
+The extracted GAF images for three datasets can be downloaded:
+[UTD-MHAD](https://drive.google.com/file/d/13Odg9lR5_nT9KWhHC7bczTss1inK6HWN/view?usp=sharing)         
+[Berkeley-MHAD](https://drive.google.com/file/d/1vun7yhIXgheiqXcNHtNF0wXVVIQsGa2E/view?usp=sharing)                
 [MMAct](https://mmact19.github.io/2019/)        
 
 ## Codes