Towards Interpreting Visual Information Processing in Vision-Language Models

TL;DR

This paper investigates how vision-language models process visual tokens, revealing their interpretability, object localization capabilities, and integration mechanisms, which enhances understanding and control of multimodal AI systems.

Contribution

It provides the first detailed analysis of visual token processing in VLMs, highlighting interpretability, object localization, and information integration mechanisms.

Findings

Object identification accuracy drops over 70% when object tokens are removed.

Visual token representations become more interpretable across layers.

Models extract object information from refined representations at the last token for prediction.

Abstract

Vision-Language Models (VLMs) are powerful tools for processing and understanding text and images. We study the processing of visual tokens in the language model component of LLaVA, a prominent VLM. Our approach focuses on analyzing the localization of object information, the evolution of visual token representations across layers, and the mechanism of integrating visual information for predictions. Through ablation studies, we demonstrated that object identification accuracy drops by over 70\% when object-specific tokens are removed. We observed that visual token representations become increasingly interpretable in the vocabulary space across layers, suggesting an alignment with textual tokens corresponding to image content. Finally, we found that the model extracts object information from these refined representations at the last token position for prediction, mirroring the process in text-only language models for factual association tasks. These findings provide crucial insights into how VLMs process and integrate visual information, bridging the gap between our understanding of language and vision models, and paving the way for more interpretable and controllable multimodal systems.