With the rapid development of 3D modeling technology, single view 3D reconstruction method based on deep learning has become an important research direction in the field of computer vision. However, the existing implicit function methods based on point learning have limitations in the use of perceptual spatial and structural information, and it is difficult to achieve high-precision 3D shape reconstruction. To solve this problem, this project proposes an implicit function g2ifu based on graph structure, which realizes high-quality single view 3D reconstruction by expanding 3D points into graph structure and introducing a priori boundary loss. The project innovatively proposed a mapping method from graph structure to implicit value, which significantly improved the accuracy of shape surface reconstruction by constructing a hypothetical 3D point expansion graph and designing a graph based boundary loss function. Experimental results show that the proposed method has achieved excellent performance on shapenet dataset, the IOU index reaches 0.632, and the chamfer-l1 distance is reduced to 0.159. The reconstruction quality of the proposed method is better than the existing methods on multiple object categories. In addition, by extending the multi view reconstruction task, the effectiveness and generalization ability of the method are further verified, which provides a new technical idea for the field of 3D reconstruction.