PHYSICALLY-INSPIRED DEEP LIGHT ESTIMATION METHOD FROM A HOMOGENEOUS-MATERIAL OBJECT FOR MIXED REALITY LIGHTING AND THE SYSTEM THEREOF

摘要

The present invention relates to a technology for estimating more accurate lighting conditions from pure light information reflected from a target object by removing material information through a physics-based deep learning network, detecting a target object of a single material included in an input image, , obtaining light reflection information by removing material information from the target object, estimating LDR (Low Dynamic Range) light information in the form of a panoramic image from the light reflection information reflected from the target object, and the LDR light information estimating high dynamic range (HDR) light information using