Paper presentations I.

Thoughts

• Motion tracking: very impressive results, but I guess this is just what you get when you chain several (high resolution) optimization models on top of good data. It is quite promising, though, and as computing power increases maybe this can head toward real-time. At that point, non-invasive motion tracking could be actually good.
• Stop motion: cool optimization problem, although seemed a bit early stage rather than ready for production work. I imagine this sort of algorithm will make its way into determining part sets to print, since that’s currently a rather manual process.
• Automatic rigging: steps made a lot of sense; going from an SDF medial surface representation to building a connectivity graph via circle packing connections, then embedding a skeleton into the graph. The embedding part was a bit opaque, since it was a bunch of hand-crafted loss functions plugged into an SVM. Looked pretty useful, so I’m surprised that even though this paper is pretty old, I haven’t seen these techniques bleed into artist-driven rigging beyond toys/teaching tools. It’s possible it lacks artistic control, but it could always be adjusted afterward and still be worthwhile. Heat diffusion for finding skin weights is basically just the heat method for geodesic distance, eh?
• Facial animation: F
• Motion graphs: interesting problem formulation, but results weren’t really that convincing. It is certainly an improvement over basic scripted/random motion, at least.
• Motion retargeting: More ~machine learning~ for retargeting. The primary technique involved a sort of neural compression to convert the detailed skeletal motion to a simplified primal skeleton, then another network to re-generate the smooth motion for the target skeleton. The results shown were very compelling, but it seemed doubtful that the technique generalizes to very divergent skeletons, since the main results/training data were all using different types of bipeds. It did at least handle lack of perfect correspondence, i.e. extra bones/joints and various size changes.