@inproceedings{rizzoli-etal-2025-civet,
    title = "{CIVET}: Systematic Evaluation of Understanding in {VLM}s",
    author = "Rizzoli, Massimo  and
      Alghisi, Simone  and
      Khomyn, Olha  and
      Roccabruna, Gabriel  and
      Mousavi, Seyed Mahed  and
      Riccardi, Giuseppe",
    editor = "Christodoulopoulos, Christos  and
      Chakraborty, Tanmoy  and
      Rose, Carolyn  and
      Peng, Violet",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2025",
    month = nov,
    year = "2025",
    address = "Suzhou, China",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/author-page-jing-yang-campinas/2025.findings-emnlp.239/",
    doi = "10.18653/v1/2025.findings-emnlp.239",
    pages = "4462--4480",
    ISBN = "979-8-89176-335-7",
    abstract = "While Vision-Language Models (VLMs) have achieved competitive performance in various tasks, their comprehension of the underlying structure and semantics of a scene remains understudied. To investigate the understanding of VLMs, we study their capability regarding object properties and relations in a controlled and interpretable manner. To this scope, we introduce CIVET, a novel and extensible framework for systemati\textbf{C} evaluat\textbf{I}on \textbf{V}ia controll\textbf{E}d s\textbf{T}imuli. CIVET addresses the lack of standardized systematic evaluation for assessing VLMs' understanding, enabling researchers to test hypotheses with statistical rigor. With CIVET, we evaluate five state-of-the-art VLMs on exhaustive sets of stimuli, free from annotation noise, dataset-specific biases, and uncontrolled scene complexity. Our findings reveal that 1) current VLMs can accurately recognize only a limited set of basic object properties; 2) their performance heavily depends on the position of the object in the scene; 3) they struggle to understand basic relations among objects. Furthermore, a comparative evaluation with human annotators reveals that VLMs still fall short of achieving human-level accuracy."
}