XR / Spatial Computing
ARS-Avatar Learns Where a Body Shadows Itself, Which Is What Relighting Usually Gets Wrong
Building an avatar that both animates and relights is hard because pose-dependent deformation, materials and light visibility are all tangled together in the source images. ARS-Avatar, posted September 23, separates them using a surfel representation, deferred shading for BRDF materials, and a differentiable ambient occlusion formulation with learnable per-body-part radii.