视觉-语言模型是理解3D场景还是仅仅 catalogue 物体？
Do Vision-Language Models Understand 3D Scenes or Just Catalogue Objects?
AI总结：
本文通过一个包含3034个样本的人工整理基准，探讨了视觉-语言模型对空间理解的深度有序遮挡、光学几何推断和体积重新安排规划能力，发现模型在重新安排可见布局时表现优异，但在遮挡和反射推断上表现较差。
摘要：
视觉-语言模型能够可靠地命名场景中的物体，但它们是否代表这些物体所处的3D布局？我们引入了一个包含3034个样本的人工整理基准，针对空间理解的三个组成部分：深度有序遮挡（通过三种独立的反事实操作化进行探测）、可见反射的光学几何推断，以及体积重新安排规划。六个前沿和开放权重的VLMs在18,204个响应上由训练注释者评分，没有使用LLM作为判断标准，揭示了明显的分离：在53-97%的准确率下，能够对可见布局进行重新安排的模型，在遮挡任务中表现不佳，仅在6-45%之间，而在反射任务中低于7%。一个具身推理模型重现了相同的模式。对Qwen3-VL-8B-Thinking的白盒分析显示，失败归因于视觉标记合并：在视觉编码器中可恢复的空间信息在标记压缩后变得不可用，只有在清洁的标记合并后激活被重新引入语言解码器后才恢复。
英文摘要：
Vision-language models reliably name objects in a scene, but do they represent the 3D layout those objects inhabit? We introduce a 3,034-sample human-curated benchmark targeting three components of spatial understanding: depth-ordered occlusion (probed via three independent counterfactual operationalisations), optical-geometry inference over visible reflections, and volumetric rearrangement planning. Six frontier and open-weight VLMs, scored by trained annotators on 18,204 responses with no LLM-as-judge, reveal a sharp dissociation: models that plan rearrangements over visible layouts at 53--97% accuracy and rarely violate collision constraints fall to 6--45% on occlusion and below 7% on reflections. An embodied-reasoning model reproduces the same profile. White-box analysis on Qwen3-VL-8B-Thinking localises the failure to the visual-token merger: spatial information recoverable throughout the vision encoder becomes inaccessible after token compression and only stabilises again when clean post-merger activations are patched into the language decoder.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
