Cross-Modal Projection in Multimodal LLMs Doesn't Really Project Visual Attributes to Textual Space
[Paper]    [GitHub]



Gaurav Verma1, Minje Choi1, Kartik Sharma1,
Jamelle Watson-Daniels2, Sejoon Oh1, and Srijan Kumar1

1Georgia Institute of Technology, 2Harvard University
    





Overview of our study: While the MLLM's domain-specific visual capability can be improved using fine-tuning strategies, the domain-specific richness of the image's post-projection representation does not improve. Results indicate that domain-specific visual attributes are predominantly modeled by the LLM parameters (whether frozen or not) and the projection does not necessarily play a role in mapping visual attributes to the LLM space. Through this study, we offer a potential reinterpretation of the role of cross-modal projections in MLLMs.


Technical Abstract

Multimodal large language models (MLLMs) like LLaVA and GPT-4(V) enable general-purpose conversations about images with the language modality. As off-the-shelf MLLMs may have limited capabilities on images from domains like dermatology and agriculture, they must be fine-tuned to unlock domain-specific applications. The prevalent architecture of current open-source MLLMs comprises two major modules: an image-language (cross-modal) projection network and a large language model. It is desirable to understand the roles of these two modules in modeling domain-specific visual attributes to inform the design of future models and streamline the interpretability efforts on the current models. To this end, via experiments on $4$ datasets and under 2 fine-tuning settings, we find that as the MLLM is fine-tuned, it indeed gains domain-specific visual capabilities, but the updates do not lead to the projection extracting relevant domain-specific visual attributes. Our results indicate that the domain-specific visual attributes are modeled by the LLM, even when only the projection is fine-tuned. Through this study, we offer a potential reinterpretation of the role of cross-modal projections in MLLM architectures.

Annotated Key Results

Use the next and previous sliders to go over the annotated version of our results and insights.


   
Image 1
Image 2
Image 3
Image 4
   

Paper and Bibtex

Cross-Modal Projection in Multimodal LLMs Doesn't Really Project Visual Attributes to Textual Space
Gaurav Verma, Minje Choi, Kartik Sharma, Jamelle Watson-Daniels, Sejoon Oh, Srijan Kumar
62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024)
Webpage: https://claws-lab.github.io/projection-in-MLLMs
Code: https://github.com/claws-lab/projection-in-MLLMs
arXiv: https://arxiv.org/abs/2402.16832
Bibtex:

@article{verma2024crossmodalprojection,
title={Cross-Modal Projection in Multimodal LLMs Doesn't Really Project Visual Attributes to Textual Space},
author={Verma, Gaurav and Choi, Minje and Sharma, Kartik and Watson-Daniels, Jamelle and Oh, Sejoon and Kumar, Srijan},
publisher={62nd Annual Meeting of the Association for Computational Linguistics (ACL)},
year={2024}
}


The template is built on top of the one build by Phillip Isola and Richard Zhang.