标题： GenSpace：空间感知图像生成的基准测试 显示英文标题

标题： GenSpace: Benchmarking Spatially-Aware Image Generation

摘要： 人类可以直观地在三维空间中构图和排列场景以供摄影。 然而，先进的AI图像生成器在从文本或图像提示创建图像时，能否具备类似的三维空间意识来规划场景？ 我们提出了GenSpace，这是一个新的基准测试和评估管道，以全面评估当前图像生成模型的空间意识。 此外，使用通用视觉-语言模型（VLM）的标准评估经常无法捕捉到详细的三维空间错误。 为了解决这一挑战，我们提出了一种专门的评估管道和指标，该方法利用多个视觉基础模型重建三维场景几何，并提供了一个更准确且与人类一致的空间准确性度量。 我们的研究结果显示，尽管AI模型可以生成视觉上吸引人的图像，并能够遵循一般指示，但在特定的三维细节方面，如物体放置、关系和测量上却存在困难。 我们总结了当前最先进的图像生成模型在空间感知方面的三个核心限制：1）物体透视理解；2）自我中心-他者中心转换；3）测量一致性，指出了提高图像生成中空间智能的方向。 显示英文摘要

摘要： Humans can intuitively compose and arrange scenes in the 3D space for photography. However, can advanced AI image generators plan scenes with similar 3D spatial awareness when creating images from text or image prompts? We present GenSpace, a novel benchmark and evaluation pipeline to comprehensively assess the spatial awareness of current image generation models. Furthermore, standard evaluations using general Vision-Language Models (VLMs) frequently fail to capture the detailed spatial errors. To handle this challenge, we propose a specialized evaluation pipeline and metric, which reconstructs 3D scene geometry using multiple visual foundation models and provides a more accurate and human-aligned metric of spatial faithfulness. Our findings show that while AI models create visually appealing images and can follow general instructions, they struggle with specific 3D details like object placement, relationships, and measurements. We summarize three core limitations in the spatial perception of current state-of-the-art image generation models: 1) Object Perspective Understanding, 2) Egocentric-Allocentric Transformation and 3) Metric Measurement Adherence, highlighting possible directions for improving spatial intelligence in image generation.