Visual Referring Expression Recognition: What Do Systems Actually Learn?

TL;DR

This paper empirically analyzes state-of-the-art referring expression recognition systems, revealing they often rely on dataset biases rather than linguistic reasoning, highlighting the need for careful data and model evaluation.

Contribution

It uncovers that current models may ignore linguistic structure and rely on shallow correlations, emphasizing the importance of analyzing what models truly learn in vision-language tasks.

Findings

Models achieve high accuracy without referring expressions, indicating reliance on dataset biases.

Predicting only object categories yields high precision, showing superficial cues dominate.

Analysis underscores the necessity of scrutinizing data construction and model reasoning.

Abstract

We present an empirical analysis of the state-of-the-art systems for referring expression recognition -- the task of identifying the object in an image referred to by a natural language expression -- with the goal of gaining insight into how these systems reason about language and vision. Surprisingly, we find strong evidence that even sophisticated and linguistically-motivated models for this task may ignore the linguistic structure, instead relying on shallow correlations introduced by unintended biases in the data selection and annotation process. For example, we show that a system trained and tested on the input image $\textit{without the input referring expression}$ can achieve a precision of 71.2% in top-2 predictions. Furthermore, a system that predicts only the object category given the input can achieve a precision of 84.2% in top-2 predictions. These surprisingly positive results for what should be deficient prediction scenarios suggest that careful analysis of what our models are learning -- and further, how our data is constructed -- is critical as we seek to make substantive progress on grounded language tasks.