diff --git a/README.md b/README.md
index bae8c61..d31bda1 100644
--- a/README.md
+++ b/README.md
@@ -32,7 +32,7 @@ Please cite our paper and kindly give a star for this repository if you use this
 
 ## Reproduce & Peformance
 
-* Due to the small number of dataset samples (especially SEM16), please tune the parameter of *--seed* for better performance.
+* Due to the small number of dataset samples (especially SEM16), the performance gap between different seeds will vary greatly, please tune the parameter of *--seed* for better performance.
 * We have provided checkpoints that are superior or equal to the performance reported in the paper. 
 * Please Run python files in [run_checkpoints](/run_checkpoints), you can use the trained model for prediction, and the model can be downloaded from [Google drives](https://drive.google.com/drive/folders/1W-UIVfHVgsLycTZdEIb4gNhGKCBW2wKo?usp=sharing).
 * We also use 5 random seeds to run the code directly without any other tuning parameters. The performance is as follows: