** Our method: The proposed co-attention augments the feature maps (for RPN) of a target image