Abstract
In recent years, neural signed distance function (SDF) has become one of the most effectiverepresentation methods for 3D models. By learning continuous SDFs in 3D space, neuralnetworks can predict the distance from a given query space point to its closest object surface,whose positive and negative signs denote inside and outside of the object, respectively.Training a specific network for each 3D model, which individually embeds its shape, canrealize compressed representation of objects by storing fewer network (and possibly latent)parameters. Consequently, reconstruction through network inference and surface recoverycan be achieved. In this paper, we propose an SDF prediction network using explicit keyspheres as input. Key spheres are extracted from the internal space of objects, whosecenters either have relatively larger SDF values (sphere radii), or are located at essentialpositions. By inputting the spatial information of multiple spheres which imply differentlocal shapes, the proposed method can significantly improve the reconstruction accuracywith a negligible storage cost. Compared to previous works, our method achieves the high-fidelity and high-compression 3D object coding and reconstruction. Experiments conductedon three datasets verify the superior performance of our method.
Abstract (translated)
URL
https://arxiv.org/abs/2201.07486