zhending/hateful-memes

仇恨模因挑战数据集是由Facebook AI创建的，旨在推动和衡量多模态推理和理解方面的进展。该数据集的主要任务是检测多模态模因中的仇恨言论。数据集包含PNG图像和对应的JSONL文件，其中JSONL文件包含每个模因的文本、图像路径和标签（0表示非仇恨，1表示仇恨）。数据集分为训练集、开发集和“可见”测试集，未来还会发布“不可见”测试集。评估指标为AUROC和准确率。数据集的标注可能存在一些错误，但开发集和测试集中的错误应该非常少。

The Hateful Memes Challenge is a dataset and benchmark created by Facebook AI to drive and measure progress on multimodal reasoning and understanding. The task focuses on detecting hate speech in multimodal memes. The dataset consists of PNG images and corresponding JSONL files, which include the text of the meme, the path to the image, and the label (0=not-hateful, 1=hateful). The dataset is divided into a training set, a development set, and a seen test set, with an unseen test set to be released later. The metrics used are AUROC and accuracy. Some misclassifications may exist in the training set, but they should be very rare in the development and seen test sets.