Neural Implicit Mapping via Nested Neighborhoods

• DOI: 10.48550/arxiv.2201.09147
• Publication Date: 2022-01-01
• Authors: Vinícius da Silva; Tiago Novello; Guilherme Schardong; Luiz Schirmer; Hélio Lopes; Luiz Velho
• Keywords: Tracing

We introduce a novel approach for rendering static and dynamic 3D neural signed distance functions (SDF) in real-time. rely on nested neighborhoods of zero-level sets SDFs, mappings between them. This framework supports animations achieves real-time performance without the use spatial data-structures. It consists three uncoupled algorithms representing steps. The multiscale sphere tracing focuses minimizing iteration time by using coarse approximations earlier iterations. normal mapping transfers details from fine SDF to surface neighborhood its set. is smooth it does not depend parametrizations. As result, can be used fetch normals discrete surfaces such as meshes skip later iterations when level sets. Finally, we propose an algorithm analytic calculation MLPs describe ways obtain sequences SDFs with algorithms.