核心概念
ANIM introduces a novel method for reconstructing accurate 3D human shapes from single-view RGB-D images with unprecedented accuracy.
要約
Recent advancements in human shape learning have shown the effectiveness of neural implicit models in generating 3D human surfaces.
ANIM addresses limitations of existing monocular approaches by incorporating depth observations to enhance reconstruction accuracy.
The model leverages multi-resolution pixel-aligned and voxel-aligned features to mitigate depth ambiguities and improve spatial relationships.
ANIM outperforms state-of-the-art methods using various input data types, showcasing high-quality reconstructions.
The introduction of ANIM-Real dataset enables fine-tuning for high-quality reconstruction from real-world captures.
統計
ANIMは、単一のRGB-D画像から正確な3D人間形状を再構築する革新的な手法を導入します。