PGlight-NeRF: advancing scene representation through physically-grounded lighting modeling

Neural radiance fields (NeRF) is an important work in the fields of novel views synthesis, representing scenes as 5D functions that model the color emitted by objects directly based on their positions and observed directions. The rendering approach of NeRF faces challenges in accurately capturing the rapid and complex variations in specular reflections on glossy surfaces. In the computer graphics, physically based rendering (PBR) facilitates the real-time rendering of realistic lighting effects. Drawing inspiration from this, we deeply analyze the reflectance render equation and propose PGlightNeRF to enhance the lighting representation of NeRF. Our study indicates that our approach can achieve training and rendering in a short timeframe, synthesizing high-quality specular reflections views.