BANF: Band-limited Neural Fields for Levels of Detail Reconstruction

arXiv:2404.13024v1 Announce Type: new
Abstract: Largely due to their implicit nature, neural fields lack a direct mechanism for filtering, as Fourier analysis from discrete signal processing is not directly applicable to these representations. Effective filtering of neural fields is critical to enable level-of-detail processing in downstream applications, and support operations that involve sampling the field on regular grids (e.g. marching cubes). Existing methods that attempt to decompose neural fields in the frequency domain either resort to heuristics or require extensive …

abstract analysis applications arxiv cs.cv eess.iv fields filtering fourier nature operations processing signal support type