动机: Recentadvances still fail to recover fine geometry and texture results fromsparse RGB inputs, especially under challenging human-object in-teractions scenarios
提出: In this paper, we propose a neural humanperformance capture and rendering system to generate both high-quality geometry and photo-realistic texture of both human andobjects under challenging interaction scenarios in arbitrary novelviews, from only sparse RGB streams
难点: complex occlusions raised by human-object interactions
方法：layer-wise scene decoupling strategy，同时考虑人体重建、物体重建与他们的联系
Occlusion-aware human reconstruction
Robust human-aware object tracking
Combines direction-aware neural blending weight learning and spatial-temporal texture completion
Method
Occlusion-aware Implicit Human Reconstruction: 从3D和2D里面提取feature