Blind to Position, Biased in Language: Probing Mid-Layer Representational Bias in Vision-Language Encoders for Zero-Shot Language-Grounded Spatial Understanding

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: An, Na Min, Kang, Inha, Lee, Minhyun, Shim, Hyunjung
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917353066856448
author An, Na Min
Kang, Inha
Lee, Minhyun
Shim, Hyunjung
author_facet An, Na Min
Kang, Inha
Lee, Minhyun
Shim, Hyunjung
contents Vision-Language Encoders (VLEs) are widely adopted as the backbone of zero-shot referring image segmentation (RIS), enabling text-guided localization without task-specific training. However, prior works underexplored the underlying biases within mid-layer representations that preserve positional and language-specific information. Through layer-wise investigation, we reveal that the conventionally used final-layer multimodal embeddings prioritize global semantic alignment, leading to two coupled consequences. First, vision embeddings exhibit weak sensitivity to positional cues. Second, multilingual text embeddings form language-dependent geometric shifts within the shared space. Motivated by these findings, we identify an underexplored pathway within VLE mid-layers to construct a spatial map, applicable for improving zero-shot RIS by 1-7 mIoU on nine RefCOCO benchmarks. Furthermore, leveraging mixed-language mid-layer embeddings yields enhanced spatial grounding accuracy (+7-8 mIoU and IoU@50), albeit with increased inference cost, and also improves performance on the zero-shot text-to-image retrieval task. Our work opens up the discussion about the effects of effective representational bias probing of VLEs for enhanced spatial grounding.
format Preprint
id arxiv_https___arxiv_org_abs_2509_23098
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Blind to Position, Biased in Language: Probing Mid-Layer Representational Bias in Vision-Language Encoders for Zero-Shot Language-Grounded Spatial Understanding
An, Na Min
Kang, Inha
Lee, Minhyun
Shim, Hyunjung
Computer Vision and Pattern Recognition
Artificial Intelligence
Vision-Language Encoders (VLEs) are widely adopted as the backbone of zero-shot referring image segmentation (RIS), enabling text-guided localization without task-specific training. However, prior works underexplored the underlying biases within mid-layer representations that preserve positional and language-specific information. Through layer-wise investigation, we reveal that the conventionally used final-layer multimodal embeddings prioritize global semantic alignment, leading to two coupled consequences. First, vision embeddings exhibit weak sensitivity to positional cues. Second, multilingual text embeddings form language-dependent geometric shifts within the shared space. Motivated by these findings, we identify an underexplored pathway within VLE mid-layers to construct a spatial map, applicable for improving zero-shot RIS by 1-7 mIoU on nine RefCOCO benchmarks. Furthermore, leveraging mixed-language mid-layer embeddings yields enhanced spatial grounding accuracy (+7-8 mIoU and IoU@50), albeit with increased inference cost, and also improves performance on the zero-shot text-to-image retrieval task. Our work opens up the discussion about the effects of effective representational bias probing of VLEs for enhanced spatial grounding.
title Blind to Position, Biased in Language: Probing Mid-Layer Representational Bias in Vision-Language Encoders for Zero-Shot Language-Grounded Spatial Understanding
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2509.23098