analiza aiqbench
This research addresses latent misalignment in equivariant representation learning by introducing Residual Latent Flow, a flow-based method that enforces group symmetry compliance to improve novel view synthesis quality. The approach offers a practical solution for geometry-aware generative models seeking enhanced interpretability and generalization under rotation groups SO(n). By correcting discrepancies between intended and actual latent transformations, the method strengthens the structural integrity of learned representations in computer vision tasks.
Equivariant Latent Alignment via Flow Matching under Group Symmetries
arXiv:2605.30705v2 Announce Type: replace-cross Abstract: Geometry-aware generative models and novel view synthesis approaches have shown strong potential in visual fidelity and consistency. In parallel, equivariant representation learning has emerged as a powerful framework for constructing latent spaces where analytically known group transformations could act directly, capturing geometric structure in data and enhancing both interpretability and generalization in novel view synthesis. However, we identify that existing approaches often suffer from latent misalignment, a discrepancy between the intended group action and the actually required transformations in the latent space. Consequently, the learned latents often fail to consistently preserve the equivariant relations imposed by the underlying group symmetry. To address this, we propose Residual Latent Flow, a flow-based framework that corrects the misaligned latents, thereby improving compliance with the underlying equivariance relation. Our comprehensive experiments show that our method significantly reduces latent misalignment and improves novel view synthesis quality, under rotation groups SO(n).
Abstract:Geometry-aware generative models and novel view synthesis approaches have shown strong potential in visual fidelity and consistency. In parallel, equivariant representation learning has emerged as a powerful framework for constructing latent spaces where analytically known group transformations could act directly, capturing geometric structure in data and enhancing both interpretability and generalization in novel view synthesis. However, we identify that existing approaches often suffer from latent misalignment, a discrepancy between the intended group action and the actually required transformations in the latent space. Consequently, the learned latents often fail to consistently preserve the equivariant relations imposed by the underlying group symmetry. To address this, we propose Residual Latent Flow, a flow-based framework that corrects the misaligned latents, thereby improving compliance with the underlying equivariance relation. Our comprehensive experiments show that our method significantly reduces latent misalignment and improves novel view synthesis quality, under rotation groups SO(n).
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2605.30705 [cs.CV]
(or arXiv:2605.30705v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2605.30705 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 243 cuvinte
Subiecte conexe
Equivariant Flow Matching for Symmetry-Breaking Bifurcation Problems
arXiv · 12 iun. 2026
Fixed-Dimensional Latent Flow for Generating Variable-Size 3D Molecules
arXiv · 9 sept. 2026
Exact equivariance, kept through training, buys zero-shot generalisation across the symmetry group
arXiv · 3 iun. 2026
Gromov-Monge Flow Matching for Equivariant Graph Generation
arXiv · 28 aug. 2026