Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Qi, Jianing, Liu, Jiawei, Tang, Hao, Zhu, Zhigang
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2503.17349
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918151946502144
author Qi, Jianing
Liu, Jiawei
Tang, Hao
Zhu, Zhigang
author_facet Qi, Jianing
Liu, Jiawei
Tang, Hao
Zhu, Zhigang
contents Vision Language Models (VLMs) excel at identifying and describing objects but often fail at spatial reasoning. We study why VLMs, such as LLaVA, underutilize spatial cues despite having positional encodings and spatially rich vision encoder features. Our analysis reveals a key imbalance: vision token embeddings have much larger norms than text tokens, suppressing LLM's position embedding. To expose this mechanism, we developed three interpretability tools: (1) the Position Sensitivity Index, which quantifies reliance on token order, (2) the Cross Modality Balance, which reveals attention head allocation patterns, and (3) a RoPE Sensitivity probe, which measures dependence on rotary positional embeddings. These tools uncover that vision tokens and system prompts dominate attention. We validated our mechanistic understanding through targeted interventions that predictably restore positional sensitivity. These findings reveal previously unknown failure modes in multimodal attention and demonstrate how interpretability analysis can guide principled improvements.
format Preprint
id arxiv_https___arxiv_org_abs_2503_17349
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Beyond Semantics: Rediscovering Spatial Awareness in Vision-Language Models
Qi, Jianing
Liu, Jiawei
Tang, Hao
Zhu, Zhigang
Computer Vision and Pattern Recognition
Vision Language Models (VLMs) excel at identifying and describing objects but often fail at spatial reasoning. We study why VLMs, such as LLaVA, underutilize spatial cues despite having positional encodings and spatially rich vision encoder features. Our analysis reveals a key imbalance: vision token embeddings have much larger norms than text tokens, suppressing LLM's position embedding. To expose this mechanism, we developed three interpretability tools: (1) the Position Sensitivity Index, which quantifies reliance on token order, (2) the Cross Modality Balance, which reveals attention head allocation patterns, and (3) a RoPE Sensitivity probe, which measures dependence on rotary positional embeddings. These tools uncover that vision tokens and system prompts dominate attention. We validated our mechanistic understanding through targeted interventions that predictably restore positional sensitivity. These findings reveal previously unknown failure modes in multimodal attention and demonstrate how interpretability analysis can guide principled improvements.
title Beyond Semantics: Rediscovering Spatial Awareness in Vision-Language Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2503.17349