Voxel Deformation-Aware Neural Intersection Function

TL;DR

This paper enhances neural intersection functions to handle deformable, animated geometry by introducing a rest-space formulation and advanced training techniques, enabling consistent and efficient dynamic geometry representation.

Contribution

It extends LSNIF with a rest-space approach and new training methods to support deformable and animated objects without retraining.

Findings

Supports parameterized deformable and animated geometry.

Maintains accuracy with deformation-aware training techniques.

Enables robust neural intersection prediction for dynamic geometry.

Abstract

We extend the Locally-Subdivided Neural Intersection Function (LSNIF) to support parameterized deformable and animated geometry. Our approach introduces a rest-space and deformed-space formulation inspired by meshless rendering, allowing ray samples to be mapped back to a canonical space where a single neural network represents geometry consistently across poses without retraining. To maintain accuracy under deformation-aware training, we incorporate scale-invariant distance regression, uncertainty-weighted multi-task learning, and a hybrid positional-grid encoding. The resulting method preserves the compactness and efficiency of LSNIF while enabling robust neural intersection prediction for dynamic geometry.