Common Objects Out of Context (COOCo): Investigating Multimodal Context and Semantic Scene Violations in Referential Communication

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Merlo, Filippo, Takmaz, Ece, Chen, Wenkai, Gatt, Albert
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917261805092864
author Merlo, Filippo
Takmaz, Ece
Chen, Wenkai
Gatt, Albert
author_facet Merlo, Filippo
Takmaz, Ece
Chen, Wenkai
Gatt, Albert
contents To what degree and under what conditions do VLMs rely on scene context when generating references to objects? To address this question, we introduce the $\textit{Common Objects Out-of-Context (COOCo)}$ dataset and conduct experiments on several VLMs under different degrees of scene-object congruency and noise. We find that models leverage scene context adaptively, depending on scene-object semantic relatedness and noise level. Based on these consistent trends across models, we turn to the question of how VLM attention patterns change as a function of target-scene semantic fit, and to what degree these patterns are predictive of categorisation accuracy. We find that successful object categorisation is associated with increased mid-layer attention to the target. We also find a non-monotonic dependency on semantic fit, with attention dropping at moderate fit and increasing for both low and high fit. These results suggest that VLMs dynamically balance local and contextual information for reference generation. Dataset and code are available here: $\href{https://github.com/cs-nlp-uu/scenereg}{https://github.com/cs-nlp-uu/scenereg}$.
format Preprint
id arxiv_https___arxiv_org_abs_2506_22274
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Common Objects Out of Context (COOCo): Investigating Multimodal Context and Semantic Scene Violations in Referential Communication
Merlo, Filippo
Takmaz, Ece
Chen, Wenkai
Gatt, Albert
Computer Vision and Pattern Recognition
Computation and Language
To what degree and under what conditions do VLMs rely on scene context when generating references to objects? To address this question, we introduce the $\textit{Common Objects Out-of-Context (COOCo)}$ dataset and conduct experiments on several VLMs under different degrees of scene-object congruency and noise. We find that models leverage scene context adaptively, depending on scene-object semantic relatedness and noise level. Based on these consistent trends across models, we turn to the question of how VLM attention patterns change as a function of target-scene semantic fit, and to what degree these patterns are predictive of categorisation accuracy. We find that successful object categorisation is associated with increased mid-layer attention to the target. We also find a non-monotonic dependency on semantic fit, with attention dropping at moderate fit and increasing for both low and high fit. These results suggest that VLMs dynamically balance local and contextual information for reference generation. Dataset and code are available here: $\href{https://github.com/cs-nlp-uu/scenereg}{https://github.com/cs-nlp-uu/scenereg}$.
title Common Objects Out of Context (COOCo): Investigating Multimodal Context and Semantic Scene Violations in Referential Communication
topic Computer Vision and Pattern Recognition
Computation and Language
url https://arxiv.org/abs/2506.22274