A Hybrid Approach for Visual Recognition of Object States

Filippos Gouidis, Katerina Papantoniou, Konstantinos Papoutsakis, Theodore Patkos, Antonis Argyros, Dimitris Plexousakis

Keywords: Objest State Recognition, Graph Neural Networks, Knowledge Graphs, Neurosymbolic Integration

Abstract: This work addresses the challenging problem of recognizing object states in a visual context by integrating data-driven and symbolic approaches. In particular, we focus on the Zero-shot variation of this task. Key contributions include the development of novel methods that exhibit state-of-the-art (SOTA) performance, the creation of a new object states dataset, the formulation of novel problems, the successful integration of low-level and high-level approaches, and comprehensive analyses that highlight the specific challenges posed by the problem.

Submission Number: 100