Hyperbolic Image Segmentation

TL;DR

This paper introduces hyperbolic manifolds for image segmentation, offering advantages like uncertainty estimation, boundary detection, zero-label generalization, and improved performance in low-dimensional embeddings.

Contribution

It proposes a novel hyperbolic space formulation for hierarchical pixel-level classification, expanding beyond traditional Euclidean methods.

Findings

Hyperbolic space improves segmentation accuracy in low-dimensional embeddings.

The method enables uncertainty estimation and boundary detection without extra labels.

Zero-label generalization is achieved through hyperbolic representations.

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 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.