Beyond Memorization: A Multi-Modal Ordinal Regression Benchmark to Expose Popularity Bias in Vision-Language Models

TL;DR

This paper reveals a popularity bias in vision-language models by introducing a large multi-modal dataset and benchmark, demonstrating models' reliance on memorization rather than understanding, especially on less popular items.

Contribution

The paper introduces the YearGuessr dataset and a new benchmark for evaluating popularity bias in vision-language models, along with popularity-aware metrics and a new model, YearCLIP.

Findings

Models perform up to 34% better on popular buildings.

VLMs struggle with less recognized, less popular items.

Benchmark reveals reliance on memorization over understanding.

Abstract

We expose a significant popularity bias in state-of-the-art vision-language models (VLMs), which achieve up to 34% higher accuracy on famous buildings compared to ordinary ones, indicating a reliance on memorization over generalizable understanding. To systematically investigate this, we introduce the largest open benchmark for this task: the YearGuessr dataset, a collection of 55,546 building images with multi-modal attributes from 157 countries, annotated with continuous ordinal labels of their construction year (1001-2024), GPS data, and page-view counts as a proxy for popularity. Using this dataset, we frame the construction year prediction task as ordinal regression and introduce popularity-aware interval accuracy metrics to quantify this bias. Our resulting benchmark of 30+ models, including our YearCLIP model, confirms that VLMs excel on popular, memorized items but struggle significantly with unrecognized subjects, exposing a critical flaw in their reasoning capabilities. Project page: https://sytwu.github.io/BeyondMemo/