NeRF as Non-Distant Environment Emitter in Physics-based Inverse Rendering

• DOI: 10.48550/arxiv.2402.04829
• Publication Date: 2024-02-07
• Authors: Jingwang Ling; Ruihan Yu; Xu Feng; Chun Du; Shuang Zhao
• Keywords: FOS: Computer and information sciences; Computer Science - Graphics; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition; Graphics (cs.GR)

Physics-based inverse rendering aims to jointly optimize shape, materials, and lighting from captured 2D images. Here is an important part of achieving faithful light transport simulation. While the environment map commonly used as model in rendering, we show that its distant assumption leads spatial invariant lighting, which can be inaccurate approximation real-world rendering. We propose use NeRF a spatially varying build pipeline using non-distant emitter. By comparing our method with on real synthetic datasets, NeRF-based emitter models scene more accurately accurate Project page video: https://nerfemitterpbir.github.io/.