PortraitCraft
收藏资源简介:
PortraitCraft是一个由CVPR 2026研讨会竞赛组织者构建的大规模人像构图数据集,旨在推动人像美学分析与可控图像合成的前沿研究。该数据集包含约50,000张经过精心筛选的真实人像图像,提供了多层次监督信息,包括全局构图评分、13个细粒度构图属性标注、属性级解释文本、视觉问答对以及用于生成任务的结构化构图描述。数据来源于真实人像摄影,通过专业标注流程创建,涵盖了主体姿态、面部表情、空间布局及背景交互等关键构图要素。该数据集主要应用于人像构图理解与生成领域,旨在解决现有数据集在结构化人像构图分析与可控生成方面的不足,为AI模型提供标准化评估平台。
PortraitCraft is a large-scale portrait composition dataset developed by the organizers of the CVPR 2026 Workshop and Competition, aimed at advancing cutting-edge research in portrait aesthetics analysis and controllable image synthesis. This dataset includes approximately 50,000 carefully curated real portrait images, and provides multi-level supervision information, including global composition scores, 13 fine-grained compositional attribute annotations, attribute-level explanatory texts, visual question-answer pairs, and structured composition descriptions for generation tasks. Derived from real portrait photography and created via a professional annotation pipeline, the dataset covers key compositional elements such as subject pose, facial expressions, spatial layout, and background interactions. Primarily utilized in the field of portrait composition understanding and generation, this dataset aims to address the limitations of existing datasets in structured portrait composition analysis and controllable generation, providing a standardized evaluation platform for AI models.

- 1The 1st PortraitCraft Challenge: A CVPR 2026 Workshop Competition on Portrait Composition Understanding and Generation · 2026年



