Title: Articulated Kinematics Distillation from Video Diffusion Models Show Chinese title

Title: 从视频扩散模型中提炼articulated运动学

Abstract: We present Articulated Kinematics Distillation (AKD), a framework for generating high-fidelity character animations by merging the strengths of skeleton-based animation and modern generative models. AKD uses a skeleton-based representation for rigged 3D assets, drastically reducing the Degrees of Freedom (DoFs) by focusing on joint-level control, which allows for efficient, consistent motion synthesis. Through Score Distillation Sampling (SDS) with pre-trained video diffusion models, AKD distills complex, articulated motions while maintaining structural integrity, overcoming challenges faced by 4D neural deformation fields in preserving shape consistency. This approach is naturally compatible with physics-based simulation, ensuring physically plausible interactions. Experiments show that AKD achieves superior 3D consistency and motion quality compared with existing works on text-to-4D generation. Project page: https://research.nvidia.com/labs/dir/akd/ Show Chinese abstract

Abstract: 我们提出了Articulated Kinematics Distillation（AKD），这是一种通过融合基于骨架的动画和现代生成模型的优势来生成高保真角色动画的框架。 AKD 使用基于骨架的表示法处理绑定的3D资源，通过专注于关节级别的控制，显著减少了自由度（DoFs），从而实现了高效且一致的运动合成。 通过使用预训练的视频扩散模型进行Score Distillation Sampling（SDS），AKD 能够蒸馏复杂的、有表现力的动作，同时保持结构完整性，克服了4D神经变形场在保持形状一致性方面面临的挑战。 这种方法天然兼容基于物理的模拟，确保了物理上合理的交互。 实验表明，与现有的文本到4D生成方法相比，AKD 在3D一致性以及运动质量方面表现出色。 项目页面：https://research.nvidia.com/labs/dir/akd/