Creative Coding
M-plicits Builds a Surface Out of Nested MLPs, and the Coarse One Filters Your Scan Noise
Implicit neural representations of surfaces force a three-way trade between training efficiency, rendering speed and noise robustness. M-plicits, posted September 23, models a surface as a residual sum of MLPs trained on a sequence of nested neighbourhoods, localising supervision to narrow bands around previous zero-level sets — which makes the coarse network act as a low-pass filter on noisy input.