DMFormer: Closing the Gap Between CNN and Vision Transformers