Spiking NeRF: Representing the Real-World Geometry by a Discontinuous Representation

• DOI: 10.1609/aaai.v38i12.29285
• Publication Date: 2024-03-25T11:06:17Z
• Authors: Zhanfeng Liao; Yan Liu; Qian Zheng; Gang Pan
• Keywords: Representation

A crucial reason for the success of existing NeRF-based methods is to build a neural density field geometry representation via multiple perceptron layers (MLPs). MLPs are continuous functions, however, real or frequently discontinuous at interface between air and surface. Such contrary brings problem unfaithful representation. To this end, paper proposes spiking NeRF, which leverages neurons hybrid Artificial Neural Network (ANN)-Spiking (SNN) framework faithful Specifically, we first demonstrate why fields will bring inaccuracy. Then, propose use field. We conduct comprehensive analysis neuron models then provide numerical relationship parameter theoretical accuracy geometry. Based on this, bounded Our method achieves SOTA performance. The source code supplementary material available https://github.com/liaozhanfeng/Spiking-NeRF.