3D Reconstruction of Point Cloud Based on Point-Move

This paper proposes an unsupervised learning framework based on the Point-Move network to address the issue of converting point cloud data to high-quality surface meshes in the field of 3D reconstruction. By utilizing an end-to-end unsupervised learning approach, the network directly learns the signed distance function (SDF) of 3D shapes from the original point cloud data, enabling a direct conversion from point cloud data to high-quality surface meshes. Experimental validation was conducted using high-quality samples from the ShapeNet and ModelNet datasets, and the accuracy was evaluated using the average distance between the point cloud and the mesh as an evaluation metric. The experimental results demonstrate that the proposed method achieves excellent performance with high precision, providing new ideas and methods for the development of 3D reconstruction technology.