In　the　single-view　reconstruction　task,　there　is　some　ambiguity　in　the　3D　shape　of　the　object　in　the　input　image.　In　the　real　world,　humans　use　the　shape　prior　knowledge　learned　by　themselves　to　reconstruct　the　complete　structure　of　the　observed　object.　Many　existing　methods　cannot　reconstruct　objects　with　complex　topological　structures　well.　In　this　paper,　we　propose　a　new　method　3D-STR　that　combines　deep　learning　models　and　structure　inference　networks.　We　use　a　deep　learning　model　with　the　sequence　structure,　which　estimates　the　2.5D　sketch　and　3D　object　shape　of　the　reconstructed　object　respectively.　In　order　to　overcome　the　ambiguity　problem　mentioned　above,　we　use　the　object　structure　inferred　from　the　input　image　to　correct　the　cumulative　errors　in　the　reconstruction　stage.　Our　method　presented　in　this　paper　achieved　the　better　results　of　loU　and　CD　on　ShapeNet　dataset.　The　experiments　indicate　the　effectiveness　of　our　proposed　method.　©　2021　IEEE.