Computing the rigid embeddings for part segmentation

[m43]

Thanks for the insightful work!

Could someone please point me toward where the rigid embeddings are computed? These embeddings are used in the reported rigid part segmentation results (Fig. 4) and were optimized with the ARAP loss (Sec. 3.3).

Does the current codebase and checkpoint support computing the rigid embeddings, or is that part of the upcoming SpatialTracker-v2 codebase?