Neural Acquisition & Representation of Subsurface Scattering
Researchers have developed a novel method for acquiring and representing subsurface scattering properties of light transport in objects. This technique utilizes a U-Net Convolutional Neural Network (CNN) that learns pixel footprint responses from 3D scan data. By employing a stereo projector-camera setup with phase-shifted profilometry patterns, the system captures detailed scattering data, enabling realistic relighting of objects with arbitrary high-resolution patterns. AI