Neural Adaptive SCEne Tracing

• DOI: 10.48550/arxiv.2202.13664
• Publication Date: 2022-01-01
• Authors: Rui Li; Darius Rückert; Yuanhao Wang; Ramzi Idoughi; Wolfgang Heidrich
• Keywords: Tracing; 3D rendering; Path tracing

Neural rendering with implicit neural networks has recently emerged as an attractive proposition for scene reconstruction, achieving excellent quality albeit at high computational cost. While the most recent generation of such methods made progress on (inference) times, very little been improving reconstruction (training) times. In this work, we present Adaptive Scene Tracing (NAScenT), first method based directly training a hybrid explicit-implicit representation. NAScenT uses hierarchical octree representation one network per leaf node and combines two-stage sampling process that concentrates ray samples where they matter near object surfaces. As result, is capable reconstructing challenging scenes including both large, sparsely populated volumes like UAV captured outdoor environments, well small geometric complexity. outperforms existing approaches in terms time.