Abstract:Implicit Neural representations (INRs) are widely used for scientific data reduction and visualization by modeling the function that maps a spatial location to a data value. Without any prior knowledge about the spatial distribution of values, we are forced to sample densely from INRs to perform visualization tasks like iso-surface extraction which can be very computationally expensive. Recently, range analysis has shown promising results in improving the efficiency of geometric queries, such as ray casting and hierarchical mesh extraction, on INRs for 3D geometries by using arithmetic rules to bound the output range of the network within a spatial region. However, the analysis bounds are often too conservative for complex scientific data. In this paper, we present an improved technique for range analysis by revisiting the arithmetic rules and analyzing the probability distribution of the network output within a spatial region. We model this distribution efficiently as a Gaussian distribution by applying the central limit theorem. Excluding low probability values, we are able to tighten the output bounds, resulting in a more accurate estimation of the value range, and hence more accurate identification of iso-surface cells and more efficient iso-surface extraction on INRs. Our approach demonstrates superior performance in terms of the iso-surface extraction time on four datasets compared to the original range analysis method and can also be generalized to other geometric query tasks.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve the efficiency of iso - surface extraction on Implicit Neural Representations (INRs). Specifically, the core of the problem lies in: 1. **Computational cost of dense sampling**: In order to perform visualization tasks (such as iso - surface extraction), it is usually necessary to densely sample data values from INRs, which is very computationally expensive. 2. **Conservatism of range analysis**: Although existing range analysis methods can provide the boundaries of the output range without dense reconstruction, these boundaries are too conservative, resulting in the inability to effectively skip uninteresting areas, thus affecting the extraction efficiency. ### Solution To solve the above problems, the author proposes a technique based on uncertainty propagation. By re - examining arithmetic rules and analyzing the probability distribution of network outputs in spatial regions, the range analysis method is improved. The specific steps are as follows: 1. **Probability distribution modeling**: Consider the input region as a source of uncertainty and propagate this uncertainty through the neural network to generate an estimated probability distribution of output values. Using the Central Limit Theorem (CLT), this distribution is efficiently modeled as a Gaussian distribution. 2. **Application of soft boundaries**: By excluding low - probability values, the output boundaries can be tightened, thereby estimating the value range more accurately. This enables more precise identification of cells containing iso - surfaces and more efficient iso - surface extraction. 3. **Adaptive query**: Using "soft" boundaries allows values to be adaptively queried in the spatial domain, thereby avoiding unnecessary dense reconstruction and improving the efficiency of iso - surface extraction. ### Main contributions - Proposed a method using uncertainty propagation techniques to efficiently estimate the output distribution of implicit neural representations on the input domain. - Demonstrated how to use this output distribution to construct a hierarchical data structure to achieve efficient iso - surface extraction without pre - processing and storing metadata. ### Related formulas - Variance calculation of Gaussian distribution: \[ \sigma^2=\frac{1}{12}(x_{\text{upper}} - x_{\text{lower}})^2 \] - Variance summation in Lindeberg CLT: \[ s_n^2=\sum_{i = 1}^{n}\sigma_i^2 \] - Distribution estimation of Probability Affine Form (PAF): \[ \hat{X}\sim N(x_0, s_N^2),\quad s_N^2=\sum_{i = 1}^{N}x_i^2 \] Through this method, the paper shows that the iso - surface extraction time on four datasets is better than the original range analysis method and can be generalized to other geometric query tasks.

Improving Efficiency of Iso-Surface Extraction on Implicit Neural Representations Using Uncertainty Propagation

Iso-Points: Optimizing Neural Implicit Surfaces with Hybrid Representations

Error-aware Sampling in Adaptive Shells for Neural Surface Reconstruction

Uncertainty modeling for fine-tuned implicit functions

Optimizing 3D Geometry Reconstruction from Implicit Neural Representations

Rendering and Polygonization of Implicit Surface by Interval Analysis Based on GPU

Estimating 3D Uncertainty Field: Quantifying Uncertainty for Neural Radiance Fields

Surface Reconstruction Based on the Modified Gauss Formula

A Comparative Study of Neural Surface Reconstruction for Scientific Visualization

HR-NeuS: Recovering High-Frequency Surface Geometry via Neural Implicit Surfaces

From Transparent to Opaque: Rethinking Neural Implicit Surfaces with $α$-NeuS

Enhancing Surface Neural Implicits with Curvature-Guided Sampling and Uncertainty-Augmented Representations

Learning to Infer Implicit Surfaces without 3D Supervision

Optimizing Implicit Neural Representations from Point Clouds via Energy-Based Models

UncertaINR: Uncertainty Quantification of End-to-End Implicit Neural Representations for Computed Tomography

A High-Performance SurfaceNets Discrete Isocontouring Algorithm

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

NeuSG: Neural Implicit Surface Reconstruction with 3D Gaussian Splatting Guidance

NeuSurf: On-Surface Priors for Neural Surface Reconstruction from Sparse Input Views

Estimation and Visualization of Isosurface Uncertainty from Linear and High-Order Interpolation Methods

Recovering Fine Details for Neural Implicit Surface Reconstruction