anaaa2/moralclip_dataset

MoralCLIP数据集是一个基于道德基础理论（MFT）的多模态数据集，用于训练和评估模型在道德内容理解方面的性能。该数据集由15000个图像-文本对组成，每个对都标注了5种不同的道德基础标签：关怀/伤害、公平/欺骗、忠诚/背叛、尊重/颠覆和圣洁/堕落。数据集分为训练集、验证集和测试集，分别包含13500、750和750个样本。图像来源于SMID、LAION-400M和ImageNet，获取方式各异，且遵循不同的许可协议。

The MoralCLIP dataset is a multimodal dataset based on Moral Foundations Theory (MFT) for training and evaluating models on moral content understanding. The dataset consists of 15,000 image-text pairs, each annotated with labels corresponding to 5 different moral foundations: Care/Harm, Fairness/Cheating, Loyalty/Betrayal, Respect/Subversion, and Sanctity/Degradation. The dataset is split into training, validation, and test sets, containing 13,500, 750, and 750 samples respectively. The images are sourced from SMID, LAION-400M, and ImageNet, each with different acquisition methods and license agreements.