List view
v0.2.2 — Data-Informed Stabilization and Scaling The v0.2.2 milestone focuses on moving the GIMBAL model from an architecturally complete but fragile system to a stable, data-informed, and collaborative modeling framework suitable for multi-person development and realistic datasets. The primary goals of this release are: Integrate Anipose-derived information (camera calibration, triangulated 3D poses, coverage statistics) as informative priors rather than treating all geometry as weakly constrained. Introduce stronger, biologically informed priors on segment lengths and selected joint positions, including support for population-level models and subject-specific measurements. Generalize anchoring mechanisms so that root position or other skeletal points can be constrained by high-quality external data when appropriate. Develop a diagnostic capability ladder that incrementally increases model and data complexity, providing a shared, reproducible basis for evaluating inference stability. Begin addressing temporal scalability through downsampled or hierarchical HMM formulations, while preserving the full-resolution pose model. v0.2.2 is explicitly a stabilization and infrastructure release: it prioritizes clarity, reproducibility, and principled use of available data over adding new model features. The outcome should be a model that samples reliably in well-defined regimes, supports collaborative development, and provides a clear path toward handling long recordings and larger datasets in future releases.
No due date•0/7 issues closed