Efficient Monocular Facial Reflectance Reconstruction via Multi-Domain Codebook Learning and Identity-Conditioned Swapping

This work presents a novel framework for monocular facial reflectance reconstruction that learns high-quality multi-domain discrete codebooks to obtain a reliable reflection prior from limited captured data, and employs identity features as conditions to reconstruct multi-view reflectance images directly from the multi-domain codebooks, which are then stitched together into a complete reflectance map.