AI RESEARCH
Structured SIR: Efficient and Expressive Importance-Weighted Inference for High-Dimensional Image Registration
arXiv CS.LG
•
ArXi:2603.17415v1 Announce Type: cross Image registration is an ill-posed dense vision task, where multiple solutions achieve similar loss values, motivating probabilistic inference. Variational inference has previously been employed to capture these distributions,. however. restrictive assumptions about the posterior form can lead to poor characterisation, overconfidence and low-quality samples. flexible posteriors are typically bottlenecked by the complexity of high-dimensional covariance matrices required for dense 3D image registration.