Showing 1–1 of 1 results for author: GhadimiAtigh, M
-
Hyperbolic Image Segmentation
Authors:
Mina GhadimiAtigh,
Julian Schoep,
Erman Acar,
Nanne van Noord,
Pascal Mettes
Abstract:
For image segmentation, the current standard is to perform pixel-level optimization and inference in Euclidean output embedding spaces through linear hyperplanes. In this work, we show that hyperbolic manifolds provide a valuable alternative for image segmentation and propose a tractable formulation of hierarchical pixel-level classification in hyperbolic space. Hyperbolic Image Segmentation opens…
▽ More
For image segmentation, the current standard is to perform pixel-level optimization and inference in Euclidean output embedding spaces through linear hyperplanes. In this work, we show that hyperbolic manifolds provide a valuable alternative for image segmentation and propose a tractable formulation of hierarchical pixel-level classification in hyperbolic space. Hyperbolic Image Segmentation opens up new possibilities and practical benefits for segmentation, such as uncertainty estimation and boundary information for free, zero-label generalization, and increased performance in low-dimensional output embeddings.
△ Less
Submitted 11 March, 2022;
originally announced March 2022.
