Timezone: »

Hyperbolic Image Segmentation
Mina Ghadimi Atigh · Julian Schoep · Erman Acar · Nanne van Noord · Pascal Mettes

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.

Author Information

Mina Ghadimi Atigh (University of Amsterdam)
Julian Schoep (Promaton)
Erman Acar (University of Amsterdam)

Erman Acar is an assistant professor for Explainable AI in Finance at the University of Amsterdam (UvA) with a diverse background of research. Prior to UvA, he has taken research positions in Knowledge Representation and Reasoning group at Vrije Universteit Amsterdam and Reinforcement Learning Group at Leiden University. Erman's current research focuses on building explainable AI architectures using causality and neuro-symbolic approaches with socially desirable guarantees and the use cases targeting financial domain with partners from both industry and the academia. Keywords: Multi-Agent Reinforcement Learning, Causal Discovery, Differentiable Logics, Forecasting.

Nanne van Noord (University of Amsterdam)
Pascal Mettes (University of Amsterdam)

More from the Same Authors