passing2961/stark-image

Stark是一个公开的大规模、长期多模态对话数据集，涵盖了多种社交角色、多模态格式、时间间隔和图像。数据集的构建使用了名为MCU的多模态上下文框架，结合了ChatGPT和Plan-and-Execute Image Aligner技术。数据集中的图像来源于多个渠道，包括个性化文本到图像生成器、Bing搜索和图像数据库检索。数据集以WebDataset格式存储，包含图像数据和相关元数据。

Stark is a publicly available, large-scale, long-term multi-modal conversation dataset that encompasses a diverse range of social personas, multi-modality formats, time intervals, and images. The dataset is automatically constructed using a novel multi-modal contextualization framework, MCU, which generates long-term multi-modal dialogues distilled from ChatGPT and the proposed Plan-and-Execute Image Aligner. The dataset contains approximately 1.72M images, stored and provided in WebDataset format. The dataset structure includes unique identifiers, image URLs, image data, and additional metadata. The dataset is constructed using a personalized text-to-image generative model, image database retrieval, and web search. The dataset is intended for research purposes only, and users should be mindful of ethical considerations when utilizing it.