ComposeHuman
收藏资源简介:
ComposeHuman数据集由中山大学、新加坡国立大学、Pixocial Technology和鹏城实验室联合创建,旨在支持多模态人类图像生成任务。该数据集包含人类图像、手绘布局、细粒度文本描述和人类组件集合,通过解耦的文本和参考图像注释,提供了更灵活的空间布局控制。数据集的应用领域主要集中在时尚设计、虚拟角色创建和社交媒体内容生成,旨在解决现有方法在复杂场景下多模态信息融合不足的问题。
The ComposeHuman Dataset was jointly developed by Sun Yat-sen University, National University of Singapore, Pixocial Technology and Peng Cheng Laboratory, with the goal of supporting multimodal human image generation tasks. This dataset encompasses human images, hand-drawn layouts, fine-grained text descriptions and human component collections, and delivers more flexible spatial layout control via decoupled text and reference image annotations. Its primary application domains include fashion design, virtual character creation and social media content generation, and it is designed to resolve the issue of insufficient multimodal information fusion in existing methods when handling complex scenarios.




