从单张2D图生成3D骨骼，能自动优化拓扑结构和几何形状。
Cortex-Synth: Differentiable Topology-Aware 3D Skeleton Synthesis with Hierarchical Graph Attention
- 分层图注意力+谱优化，实现骨骼结构与拓扑的联合生成
- 在ShapeNet上关键点误差降低18.7%，拓扑错误减少42%
- 适合机器人操控、医学影像和角色绑定等需精确骨骼的应用
我们提出Cortex-Synth，一个端到端可微的框架，仅凭单张2D图像即可联合生成3D骨骼的几何结构与拓扑。该架构引入三项创新：(1) 多尺度骨骼细化的分层图注意力机制，(2) 基于拉普拉斯特征分解的可微谱拓扑优化，(3) 用于姿态结构对齐的对抗性几何一致性训练。框架包含四个协同模块：伪3D点云生成器、增强型PointNet编码器、骨骼坐标解码器，以及新型可微图构建网络（DGCN）。实验表明，在ShapeNet上，该方法在MPJPE上提升18.7%，图编辑距离提升27.3%，拓扑错误较此前方法降低42%。其端到端可微特性使其适用于机器人操作、医学成像和自动化角色绑定等场景。
原文摘要 · Abstract (English)
We present Cortex Synth, a novel end-to-end differentiable framework for joint 3D skeleton geometry and topology synthesis from single 2D images. Our architecture introduces three key innovations: (1) A hierarchical graph attention mechanism with multi-scale skeletal refinement, (2) Differentiable spectral topology optimization via Laplacian eigen decomposition, and (3) Adversarial geometric consistency training for pose structure alignment. The framework integrates four synergistic modules: a pseudo 3D point cloud generator, an enhanced PointNet encoder, a skeleton coordinate decoder, and a novel Differentiable Graph Construction Network (DGCN). Our experiments demonstrate state-of-the-art results with 18.7 percent improvement in MPJPE and 27.3 percent in Graph Edit Distance on ShapeNet, while reducing topological errors by 42 percent compared to previous approaches. The model's end-to-end differentiability enables applications in robotic manipulation, medical imaging, and automated character rigging.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。