DAMO-NLP-SG/CMM
收藏Hugging Face2025-05-15 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/DAMO-NLP-SG/CMM
下载链接
链接失效反馈官方服务：
资源简介：
多模态诅咒（CMM）数据集是一个经过精心策划的基准，旨在评估大型多模态模型（LMMs）在视觉、音频和语言模态中产生的幻觉漏洞。该数据集包含了来自WebVid、AudioCaps、Auto-ACD和YouTube的1200个精心挑选的视频/音频/视频-音频样本，以及针对每个样本的2400个探针问题。这些问题旨在评估模型对真实存在和非存在物体或事件的感知准确性和抗幻觉能力。
The Curse of Multi-Modalities (CMM) Dataset is a curated benchmark designed to evaluate hallucination vulnerabilities in Large Multi-Modal Models (LMMs) across visual, audio, and language modalities. The dataset consists of 1,200 carefully selected video/audio/video-audio samples from WebVid, AudioCaps, Auto-ACD, and YouTube, paired with 2,400 probing questions targeting both real existent and non-existent objects or events to assess perception accuracy and hallucination resistance of the models.
提供机构：
DAMO-NLP-SG
