MoSAR: Monocular Semi-Supervised Model for Avatar Reconstruction using Differentiable Shading | Read Paper on Bytez