Abstract:Implicit 3D surface reconstruction of an object from its partial and noisy 3D point cloud scan is the classical geometry processing and 3D computer vision problem. In the literature, various 3D shape representations have been developed, differing in memory efficiency and shape retrieval effectiveness, such as volumetric, parametric, and implicit surfaces. Radial basis functions provide memory-efficient parameterization of the implicit surface. However, we show that training a neural network using the mean squared error between the ground-truth implicit surface and the linear basis-based implicit surfaces does not converge to the global solution. In this work, we propose locally supported compact radial basis functions for a linear representation of the implicit surface. This representation enables us to generate 3D shapes with arbitrary topologies at any resolution due to their continuous nature. We then propose a neural network architecture for learning the linear implicit shape representation of the 3D surface of an object. We learn linear implicit shapes within a supervised learning framework using ground truth Signed-Distance Field (SDF) data for guidance. The classical strategies face difficulties in finding linear implicit shapes from a given 3D point cloud due to numerical issues (requires solving inverse of a large matrix) in basis and query point selection. The proposed approach achieves better Chamfer distance and comparable F-score than the state-of-the-art approach on the benchmark dataset. We also show the effectiveness of the proposed approach by using it for the 3D shape completion task.

What problem does this paper attempt to address?

The paper aims to address the problem of reconstructing implicit 3D surfaces from partial and noisy 3D point cloud scans. Specifically, the authors propose a Linear Implicit Surface Representation (LISR) method based on Compactly Supported Radial Basis Functions (CSRBF). Traditional radial basis functions perform well in parameterizing complex implicit surfaces, but methods that directly use neural networks to train and predict Signed-Distance Fields (SDF) do not converge to a global optimal solution. To overcome this challenge, the authors propose a new strategy by selecting appropriate basis functions and support points. The LISR method has the following advantages: 1. **Memory Efficiency**: Compared to neural implicit representations (such as the work by Park et al.) and algebraic implicit representations (such as the work by Yavartanoo et al.), the LISR model significantly reduces the number of parameters, making it more suitable for applications requiring efficient memory usage. 2. **Constraint Conditions**: The authors identify a constraint condition that the selection of basis functions and support points must meet specific requirements to facilitate the effective learning of linear implicit surfaces. Experiments validate that the proposed CSRBF-based method, combined with the proposed support point selection strategy, can meet this constraint condition. 3. **Neural Network Architecture**: The paper proposes a neural network architecture to learn the linear combination coefficients of CSRBF basis functions and guides the learning process through a supervised learning framework using real signed-distance field data for supervision. Additionally, the authors demonstrate the effectiveness of LISR in shape completion tasks and show through experiments that it achieves better Chamfer distance and comparable F-scores on benchmark datasets compared to existing techniques. In summary, LISR not only can generate 3D shapes of arbitrary topology but also can represent them at any resolution, thanks to its continuous nature.

LISR: Learning Linear 3D Implicit Surface Representation Using Compactly Supported Radial Basis Functions

Learning to Infer Implicit Surfaces without 3D Supervision

Deep Local Shapes: Learning Local SDF Priors for Detailed 3D Reconstruction

Local Deep Implicit Functions for 3D Shape

Implicit Filtering for Learning Neural Signed Distance Functions from 3D Point Clouds

A two-level approach to implicit surface modeling with compactly supported radial basis functions

Neural-IMLS: Self-supervised Implicit Moving Least-Squares Network for Surface Reconstruction

Neural Implicit Surface Reconstruction of Freehand 3D Ultrasound Volume with Geometric Constraints

Neural Vector Fields for Implicit Surface Representation and Inference

Neural Unsigned Distance Fields for Implicit Function Learning

Learning Neural Implicit Representations with Surface Signal Parameterizations

Normal-guided Detail-Preserving Neural Implicit Functions for High-Fidelity 3D Surface Reconstruction

Latent Partition Implicit with Surface Codes for 3D Representation

Efficient Implicit Neural Reconstruction Using LiDAR

NeuralIndicator: Implicit Surface Reconstruction from Neural Indicator Priors

Optimizing 3D Geometry Reconstruction from Implicit Neural Representations

Deep Implicit Moving Least-Squares Functions for 3D Reconstruction

Deep Learning 3d Shape Surfaces Using Geometry Images

Shape Reconstruction of Object-Level Building From Single Image Based on Implicit Representation Network

Learning Unsigned Distance Fields from Local Shape Functions for 3D Surface Reconstruction

SDF-SRN: Learning Signed Distance 3D Object Reconstruction from Static Images