视觉语言模型（VLMs）越来越多地用于分析和推理图像，但目前尚不清楚它们的内部表征是否忠实地编码了此类推理所需的结构化关系。我们引入EICS（Evaluation-Invariant Concept Stability，评估不变概念稳定性），这是一种用于测试表征在受控单概念干预下是否一致编码概念的指标；同时引入CASP-VL（Causally controlled Spatial Perturbations for Vision-Language models，面向视觉语言模型的因果控制空间扰动），这是一个源自Visual Genome场景图的因果控制空间扰动基准。在多个VLM架构上，我们发现空间关系的表征忠实度远低于词汇信息，即便相同模型在组合基准上的得分远高于随机水平。进一步分析表明，该关系在谓词自身的token处可被识别，但当句子通过（EOS）token或平均池化聚合为全局表征时，该关系会消失。这些发现揭示了一种基于输出的评估无法察觉的系统性缺陷：VLMs在下游使用的表征中未保留其输入中存在的关系信息。
英文摘要：
Vision-Language Models (VLMs) increasingly analyze and reason about images, yet it remains unclear whether their internal representations faithfully encode the structured relations required for such reasoning. We introduce EICS (Evaluation-Invariant Concept Stability), a metric for testing whether representations consistently encode concepts under controlled one-concept interventions, and CASP-VL (Causally controlled Spatial Perturbations for Vision-Language models), a benchmark of causally controlled spatial perturbations derived from Visual Genome scene graphs. Across multiple VLM architectures, we find that spatial relations are represented substantially less faithfully than lexical information, even when the same models score well above chance on compositional benchmarks. Further analysis shows that the relation is identifiable at the predicate's own tokens but is lost when the sentence is aggregated into a global representation, whether through the (EOS) token or by mean pooling. These findings reveal a systematic failure invisible to output-based evaluation: VLMs do not preserve, in the representation used downstream, relational information that is present in their own input.
VTCBench: Can Vision-Language Models Understand Long Context with Vision-Text Compression?
1 Institute of Automation, Chinese Academy of Sciences；2 School of Artificial Intelligence, University of Chinese Academy of Sciences；3 Centre for Artificial Intelligence；Robotics, Hong Kong Institute of Science \& Innovation, CAS；4 Independent Researcher
A Pilot Study on Curator-Guided Multilingual Art Description for Blind and Low-Vision Audiences with Small Vision-Language Models
Technical University of Munich（慕尼黑技术大学）；Foundation for Research and Technology -- Hellas（希腊研究与技术基金会）；National University of Science and Technology Politehnica Bucharest（布加勒斯特政治技术科学与技术国家大学）
Vision-and-Language Training Helps Deploy Taxonomic Knowledge but Does Not Fundamentally Alter It
Boston University（波士顿大学）；University of Amsterdam（阿姆斯特丹大学）；Umeå University（乌梅大学）；Vrije Universiteit Amsterdam（阿姆斯特丹自由大学）；TTIC（TTIC研究中心）；The University of Texas at Austin（德克萨斯大学奥斯汀分校）
