Ryann829/Scone-S2I-57K

Scone-S2I-57K是一个专为主题驱动图像生成任务设计的综合数据集，包含22K精炼的单候选数据和35K多候选数据，用于训练Scone模型。单候选数据中每个参考图像仅包含一个候选主题，而多候选数据中每个参考图像包含一个或多个候选主题。数据集通过视觉语言模型筛选和两步解耦策略构建，包含来自开源数据集和合成数据的不同来源。

Scone-S2I-57K is a comprehensive dataset designed for the subject-driven image generation task, consisting of 22K refined single-candidate data and 35K multi-candidate data, which are used for training the Scone model. Single-candidate data contains only one candidate subject per reference image, while multi-candidate data contains one or more candidate subjects per reference image. The dataset is curated through VLM filtering and a two-step decoupling strategy, incorporating samples from open-source datasets and synthetic data.