HAIG-2.9M
收藏资源简介:
HAIG-2.9M是一个大规模、高质量、高度多样化的数据集,用于基于关键点的图像生成,适用于人类和动物。该数据集包括786K张图像,覆盖31个物种类别,包含2.9M个实例级别的边界框、关键点和标题。数据集由爬取和过滤多个高质量数据集和网站获得,并使用最先进的模型进行注释。HAIG-2.9M旨在解决现有数据集在多类、多实例图像生成方面的局限性,为图像生成领域提供高质量的数据支持。
HAIG-2.9M is a large-scale, high-quality, and highly diverse dataset designed for keypoint-based image generation for both humans and animals. It comprises 786K images covering 31 animal categories, alongside 2.9M instance-level bounding boxes, keypoints, and captions. The dataset is developed by crawling and filtering multiple high-quality datasets and websites, and annotated using state-of-the-art models. HAIG-2.9M is intended to address the limitations of existing datasets in multi-category and multi-instance image generation, providing high-quality data support for the image generation research domain.

- 1UniMC: Taming Diffusion Transformer for Unified Keypoint-Guided Multi-Class Image Generation香港科技大学, 腾讯, 北京大学, 香港中文大学, 香港大学, 新加坡国立大学 · 2025年



