Abstract: Recently masked autoencoder (MAE) has achieved great success in visual representation learning and delivered promising potential in many downstream vision tasks. However, due to the lack of ...