Voxel Deformation-Aware Neural Intersection Function
It addresses the problem of representing dynamic, deformable geometry with neural implicit functions for rendering applications, but the gains are incremental over prior work.
This work extends LSNIF to handle deformable and animated geometry by mapping ray samples to a canonical rest space, enabling a single neural network to represent geometry across poses without retraining. The method preserves compactness and efficiency while achieving robust neural intersection prediction for dynamic shapes.
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.