DIG: Draping Implicit Garment over the Human Body

Existing data-driven methods for draping garments over human bodies, despite being effective, cannot handle of arbitrary topology and are typically not end-to-end differentiable. To address these limitations, we propose an differentiable pipeline that represents using implicit surfaces learns a skinning field conditioned on shape pose parameters articulated body model. limit body-garment interpenetrations artifacts, interpenetration-aware pre-processing strategy training data novel loss penalizes self-intersections while garments. We demonstrate our method yields more accurate results garment reconstruction deformation with respect to state the art methods. Furthermore, show method, thanks its differentiability, allows recover jointly from image observations, something previous work could do. Our code is available at https://github.com/liren2515/DIG .