Group-based Distinctive Image Captioning with Memory Attention