In this project, we introduce a memory-augumented method, which extends an existing image captioner by incorporating explict knowledge from a memory bank.