핵심 개념
SelfPIFu proposes a self-supervised framework using depth-guided learning for accurate 3D human reconstruction.
초록
PIFu and PIFuHD are state-of-the-art methods for 3D human digitization.
SelfPIFu introduces a self-supervised network named SelfPIFu for improved reconstructions.
Depth-guided learning enhances reconstruction quality on synthetic and real-world images.
SelfPIFu outperforms existing methods in terms of fidelity and detail preservation.
The framework consists of normal and depth estimators, along with a novel self-supervised SDF-based PIFu module.
Extensive experiments validate the effectiveness and superiority of SelfPIFu.
통계
합성 데이터에서 IoU는 89.03%로 PIFuHD 및 ECON보다 높음.
깊이 맵을 사용한 경우, 실제 이미지에서 기하학적 세부 사항 재구성에 우수함.
인용구
"SelfPIFu는 깊이 지도를 중간 입력으로 사용하여 3D 인간 재구성의 품질을 향상시킵니다."
"실험 결과는 SelfPIFu의 효과와 우수성을 검증합니다."