Overview
TiNeuVox is a radiance field framework utilizing time-aware voxel features for fast dynamic scene reconstruction. It introduces a coordinate deformation network to model coarse motion trajectories and enhances temporal information within the radiance network. The core innovation is a multi-distance interpolation method applied to voxel features, adept at handling both small and large motions. This method significantly accelerates the optimization process for dynamic radiance fields while preserving high rendering quality. Empirical results show training completion in approximately 8 minutes with an 8-MB storage footprint, outperforming previous dynamic NeRF methods in rendering performance. TiNeuVox leverages PyTorch, CUDA, and is built upon DirectVoxGo and D-NeRF architectures.