Affinity maximization learning for unsupervised deep visual graph matching

doi:10.1016/j.patcog.2026.113879

Affinity maximization learning for unsupervised deep visual graph matching

2026-05-07

0

PRE

AI

DOI：10.1016/j.patcog.2026.113879

Original

Original request for help

Save

Abstract

En 中文

• An unsupervised deep visual graph matching framework based on affinity maximization is proposed. • A versatile affinity-maximization loss is introduced to train without ground-truth or pseudo labels. • The framework is model-agnostic and applies to supervised and unsupervised affinity-based methods. • The framework is further extended to multi-graph matching. • Extensive experiments demonstrate strong performance on standard benchmarks.

Keywords:

unsupervised deep visual graph matching

affinity maximization

graph matching

deep learning

model-agnostic

Affinity maximization learning for unsupervised deep visual graph matching

Abstract

Journal

Organization