Animatable NeRF Dynamic Detail Enhancement Based on Residual Deformation Field with Progressive Training

We present a novel animatable neural radiance field (Animatable NeRF) framework incorporating a residual deformation field (RDF) to address the limitations of current monocular video-driven human reconstruction techniques. While existing Animatable NeRF approaches demonstrate promising results, they exhibit significant constraints in capturing fine-grained dynamic details, particularly in clothing wrinkles and hair motion. To overcome these limitations, we propose a comprehensive solution featuring two key innovations: 1) Residual Deformation Field (RDF): Predicts local displacement residuals in canonical space, effectively decoupling global pose from dynamic details; 2) Progressive Training: A staged optimization strategy that prevents local details from interfering with the learning of global deformations. Experimental evaluations demonstrate that our framework advances human geometry and appearance reconstruction capabilities, enables high-fidelity novel-view synthesis, and supports robust human animation under novel poses.