Unsupervised Label Learning on Manifolds by Spatially Regularized Geometric Assignment
Manifold models of image features abound in computer vision. We present a novel approach that combines unsupervised computation of representative manifold-valued features, called labels, and the spatially regularized assignment of these labels to given manifold-valued data. Both processes evolve dynamically through two Riemannian gradient flows that are coupled. The representation of labels and assignment variables are kept separate, to enable the flexible application to various manifold data models. As a case study, we apply our approach to the unsupervised learning of covariance descriptors on the positive definite matrix manifold, through spatially regularized geometric assignment.
This work was supported by the German Research Foundation (DFG), grant GRK 1653.
- 5.Cherian, A., Sra, S.: Positive definite matrices: data representation and applications to computer vision. In: Minh, H.Q., Murino, V. (eds.) Algorithmic Advances in Riemannian Geometry and Applications. ACVPR, pp. 93–114. Springer, Cham (2016). https://doi.org/10.1007/978-3-319-45026-1_4CrossRefzbMATHGoogle Scholar
- 15.Savarino, F., Hühnerbein, R., Åström, F., Recknagel, J., Schnörr, C.: Numerical integration of Riemannian gradient flows for image labeling. In: Lauze, F., Dong, Y., Dahl, A.B. (eds.) SSVM 2017. LNCS, vol. 10302, pp. 361–372. Springer, Cham (2017). https://doi.org/10.1007/978-3-319-58771-4_29CrossRefGoogle Scholar
- 16.Sra, S.: Positive Definite Matrices and the Symmetric Stein Divergence. CoRR abs/1110.1773 (2013)Google Scholar