EunsuKim/mix-culture-food-train

混合文化美食数据集训练集包含5798个多模态训练样本，遵循Swift的“消息+图像”模式。数据集旨在用于细调项目，保留了所有2_synt_final_0914组合以保持food2多样性，其他合成子集则按基础ID大致减少了三分之一，同时保持背景覆盖平衡。在使用前需要指定许可和再分发条款。

The Mix Culture Food Dataset training split includes 5,768 multimodal training samples following the Swift messages + images schema. The dataset is designed for fine-tuning projects, retaining all 2_synt_final_0914 combinations to preserve food2 diversity, while other synthetic subsets are down-sampled to roughly one third per base ID, maintaining balanced background coverage. License and redistribution terms need to be specified before public release.