TL;DR: DragMesh enables real time, physically valid 3D object articulation by decoupling kinematic reasoning from motion generation and producing plausible motions via a dual quaternion based generative model.





4D Visualization





Efficiency


You browser does not support this image.

Bubble size indicates parameter count (core module only), color denotes generalization capability (Purple: generalizable; Orange: per-object training required). Our method achieves generalization with significantly lower computational cost than existing generalizable approaches.





DragMesh Pipeline


You browser does not support this image.

The DragMesh pipeline fuses point cloud, joint, and drag inputs through a VAE and Transformer architecture to predict a physically-corrected dual quaternion.





DregMesh Inference Pipeline


You browser does not support this image.

The DragMesh Annotation-Free Inference Pipeline. Given a raw mesh and drag, a segmentation model identifies the movable part while a VLM predicts the joint type. Our KPP-Net then regresses the precise axis and origin, enabling the final DragMesh model to generate the Dual Quaternion animation.