Skeleton?aware Implicit Function for single?view human reconstruction

The aim is to reconstruct a complete and detailed clothed human from single-view input. Implicit function suitable for this task because it represents fine shape details varied topology. Current methods, however, often suffer artefacts such as broken or disembodied body parts, missing details, depth ambiguity due the complexity of articulation. main issue observed by authors structure-agnostic. To address these problems, fully utilise skinned multi-person linear (SMPL) model propose method using Skeleton-aware Function (SIF). alleviate proposed skeleton-aware structure prior makes skeleton awareness into an implicit function, which consists bone-guided sampling strategy skeleton-relative encoding strategy. deal with authors’ body-guided pixel-aligned feature exploits SMPL enhance 2D normal semantic features, aggregation uses extra geometry-aware enabling more plausible merging less noisy geometry. Additionally, SIF also adapted RGB-D input, experimental results show that outperforms state-of-the-arts methods on challenging datasets Twindom Thuman3.0.