Excuse me, I read your paper and it's great for video tasks. But I wonder if it works for image tasks?