Scalable, Detailed and Mask-Free Universal Photometric Stereo

Surface normal map SDM-UniPS (Ours) Input image (1st out of 9) UniPS [Ikehata2022] 3D Scanner (EinScan-SE) INPUT: Multiple images under varying unknown lighting conditions 1st 2nd 23rd Figure 1. Given multiple images under unknown spatially-varying illuminations, our method can recover the detailed surface normal map of non-convex, non-Lambertian surfaces (Left). Our method even surpasses the level of detail provided by consumer 3-D scanners (Right).