Review for NeurIPS paper: RelationNet++: Bridging Visual Representations for Object Detection via Transformer Decoder