Abstract Implicit surface reconstruction based on point clouds is usually implemented by the method of SDF value estimation. It represents the surface as the zero isosurface of the implicit function and uses 3D points with supervised values near the surface to fit the surface. In order to improve the reconstruction quality, this paper makes improvements based on the framework of the Neural-IMLS algorithm and proposes an adaptive radius adjustment mechanism based on the local density of point clouds. Through the local density perception mechanism based on KNN, the ball query radius is dynamically adjusted to adapt to the characteristics of different input point clouds. Sample out more reasonable data for the implicit neural network to conduct SDF estimation. In the subsequent surface fitting process, it can enhance the recovery ability for complex local details and improve the reconstruction accuracy. To verify the effectiveness of the improved algorithm, this paper adopts the SAL algorithm, the Neural-IMLS algorithm and the algorithm in this paper to compare the implicit reconstruction of the Stanford standard point cloud model. The experimental results show that for the Armadillo, Bunny and Happy models, the CD distance and HD distance reconstructed by the improved algorithm in this paper have decreased. Among them, the CD distance of the Bunny model has decreased by 27.3%, the HD distance has decreased by 53.1%, and the CD distance of the Armadillo model has decreased by 16.6%. It decreased by 55.3% at the HD distance. Meanwhile, by qualitatively comparing its visualization results with the GT network, the reconstruction effects of each model have also improved, proving the effectiveness of the improvement points.
Zhao et al. (Mon,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: