Visual Spatial Reasoning

TL;DR

This paper introduces a new dataset for visual spatial reasoning that highlights the challenges current vision-and-language models face in understanding complex spatial relations, revealing significant performance gaps.

Contribution

The paper presents VSR, a novel dataset with diverse spatial relations and linguistic phenomena, and evaluates model limitations in capturing relational information.

Findings

Models achieve only around 70% accuracy compared to over 95% human performance.

Performance on specific relations does not correlate with training data size.

Models struggle with orientation-based spatial relations.

Abstract

Spatial relations are a basic part of human cognition. However, they are expressed in natural language in a variety of ways, and previous work has suggested that current vision-and-language models (VLMs) struggle to capture relational information. In this paper, we present Visual Spatial Reasoning (VSR), a dataset containing more than 10k natural text-image pairs with 66 types of spatial relations in English (such as: under, in front of, and facing). While using a seemingly simple annotation format, we show how the dataset includes challenging linguistic phenomena, such as varying reference frames. We demonstrate a large gap between human and model performance: the human ceiling is above 95%, while state-of-the-art models only achieve around 70%. We observe that VLMs' by-relation performances have little correlation with the number of training examples and the tested models are in general incapable of recognising relations concerning the orientations of objects.