Localization vs. Semantics: Visual Representations in Unimodal and Multimodal Models

TL;DR

This paper compares visual representations in vision-and-language models and vision-only models, revealing differences in their strengths across various tasks and providing insights into the role of language in visual learning.

Contribution

It offers a comprehensive analysis of how multimodal and unimodal models differ in visual representation quality across diverse tasks.

Findings

Vision-and-language models excel at label prediction tasks.

Vision-only models outperform in dense, localized prediction tasks.

The study provides empirical insights into the role of language in visual learning.

Abstract

Despite the impressive advancements achieved through vision-and-language pretraining, it remains unclear whether this joint learning paradigm can help understand each individual modality. In this work, we conduct a comparative analysis of the visual representations in existing vision-and-language models and vision-only models by probing a broad range of tasks, aiming to assess the quality of the learned representations in a nuanced manner. Interestingly, our empirical observations suggest that vision-and-language models are better at label prediction tasks like object and attribute prediction, while vision-only models are stronger at dense prediction tasks that require more localized information. We hope our study sheds light on the role of language in visual learning, and serves as an empirical guide for various pretrained models. Code will be released at https://github.com/Lizw14/visual_probing