Spotlights: Probing Shapes from Spherical Viewpoints

Recent years have witnessed the surge of learned representations that directly build upon point clouds. Inspired by spherical multi-view scanners, we propose a novel sampling model called Spotlights to represent 3D shape as compact 1D array depth values. It simulates configuration cameras evenly distributed on sphere, where each virtual camera casts light rays from its principal probe for possible intersections with object surrounded sphere. The structured cloud is hence given implicitly function depths. We provide detailed geometric analysis this new scheme and prove effectiveness in context completion task. Experimental results both synthetic real dataset demonstrate our method achieves competitive accuracy consistency while at lower computational cost. code will be released https://github.com/goldoak/Spotlights.