Localization vs. Semantics: Visual Representations in Unimodal and Multimodal Models
This provides an empirical guide for selecting pretrained models based on task requirements, but it is incremental as it analyzes existing models without introducing new methods.
The study compared visual representations in vision-and-language models and vision-only models, finding that vision-and-language models excel at object and attribute prediction tasks, while vision-only models are stronger at dense prediction tasks requiring localized information.
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