Results
Non-rigging
Temporally Consistent Mesh Sequence Generation
We qualitatively compared our temporally consistent mesh generation method with ActionMesh’s feed-forward video-to-mesh sequence generation, which is mostly trained on simple cases, on a video case where the shape/topology changes a lot over time. ActionMesh’s results have obvious artifacts in most frames, while our method successfully generated high-quality meshes and maintained temporal consistency.
Rigging
For the first semester, we have constructed an interactive deterministic Rig Editor for skeleton editing, validation, and export. It includes core editing tools: MOVE, REMOVE, INSERT ON EDGE, and ADD LEAF. This video shows the operations of the app.
