The　feature　pyramid　network　(FPN)　has　achieved　impressive　results　in　the　field　of　object　detection　and　instance　segmentation　by　aggregating　features　of　different　scales,　especially　the　detection　of　small　objects.　However,　for　some　special　large　objects　(such　as　tables,　chairs,　etc.),　It　is　difficult　to　achieve　good　results　for　FPN.　In　this　paper,　we　propose　a　new　simple　but　effective　network,　the　Bidirectional　Path　Network　(BPN),　for　the　problems　that　FPN　cannot　solve.　In　simple　terms,　it　consists　of　a　top-to-down　FPN　and　bottom-to-up　FPN.　This　bidirectional　network　structure　can　greatly　enrich　high-level　semantic　information　and　improve　the　detection　effect　of　these　large　objects.　And　we　also　introduce　dense　connections　to　enrich　the　output　features　further.　We　tested　our　method　on　the　COCO　dataset.　Firstly,　on　the　object　detection　task,　our　method　obtains　comparable　results　with　the　state-of-the-art　benchmark.　Then,　on　the　instance　segmentation　task,　our　method　also　achieved　good　results.　©　2021,　Springer　Nature　Switzerland　AG.