paper-with-me

홈 › Papers

Prompt Sensitivity in Vision-Language Grounding: How Small Changes in Wording Affect Object Detection

2026-04-18 · Dawar Jyoti Deka, Amit Sethi, Syed Mohammad Ali arxiv

Vision-language models enable open-vocabulary object grounding through natural language queries, under the implicit assumption that semantically equivalent descriptions yield consistent outputs. We examine this assumption using a controlled pipeline combining DETR for object proposals with CLIP for language-conditioned selection on 263 COCO val2017 images. We find that overlapping prompts such as "a person," "a human," and "a pedestrian" frequently select different instances, with mean instability of 2.11 distinct selections across six prompts. PCA analysis shows this variability is structured and directional, not random. Prompt ensembling does not improve quality and often shifts selections toward generic regions. We further show that text embedding proximity explains only 34% of grounding disagreement (r = -0.58), confirming that instability arises from the argmax selection mechanism rather than text-level distances alone.

📄 PDF Abstract BibTeX arXiv:2604.17126

Code (0)

등록된 구현이 없습니다.

Tasks

Natural Language QueriesObject Detection

Similar Papers 제목 키워드 기반

Grounded Concreteness: Human-Like Concreteness Sensitivity in Vision-Language Models

2026-01-26 · Aryan Roy, Zekun Wang, Christopher J. MacLellan arxiv

Do vision--language models (VLMs) develop more human-like sensitivity to linguistic concreteness than text-only large language models (LLMs) when both are evaluated with text-only prompts? We study this question with a c…

SPARK-SAM: Learning How to Prompt and Respond for Infrared Small Target Segmentation

2026-08-21 · Aji Mao, Zhenming Peng, Bailin Mu, Tian Pu arxiv

Promptable segmentation models provide a reusable interface, but direct transfer to automatic infrared small-target segmentation (IRSTD) exposes a mismatch between spatial prompts and target-domain mask responses. In a d…

WildRoadBench: A Wild Aerial Road-Damage Grounding Benchmark for Vision-Language Models and Autonomous Agents

2026-05-19 · Bingnan Liu, Chenhang Cui, Rui Huang, Jiani Luo 외 arxiv

We introduce WildRoadBench, a wild aerial road-damage grounding benchmark that couples direct visual grounding by vision-language models with autonomous research-and-engineering by LLM-driven agents on a single professio…

Visual Grounding

SteerSeg: Attention Steering for Reasoning Video Segmentation

2026-05-14 · Ali Cheraghian, Hamidreza Dastmalchi, Abdelwahed Khamis, Morteza Saberi 외 arxiv

Video reasoning segmentation requires localizing objects across video frames from natural language expressions, often involving spatial reasoning and implicit references. Recent approaches leverage frozen large vision-la…

Video SegmentationSpatial ReasoningText Generation

PARC: A Quantitative Framework Uncovering the Symmetries within Vision Language Models

2025-01-01 · CVPR 2025 1 · Jenny Schmalfuss, Nadine Chang, Vibashan VS, Maying Shen 외

Vision language models (VLMs) respond to user-crafted text prompts and visual inputs, and are applied to numerous real-world problems. VLMs integrate visual modalities with large language models (LLMs), which are wel…

Sensitivity