Using Relations To Describe Three-Dimensional Scenes: A Model of Spatial Relation Apprehension and Interference

Abstract

Understanding spatial information between the objects visible in a scene is crucial to tasks such as describing relevant features of the scene, navigating it based on a description of its features, and for creating novel imagined scenes based on linguistic input. We present a model of spatial relation apprehension able to map geometric information from 3D scenes containing multiple objects to English prepositions and verbs in terms of direction and distance, as well as topological relations. We created an implementation of the model to evaluate its effectiveness.


Back to Table of Contents