Self-Reviewed Emotion (SRE) Dataset; Human-Reviewed Emotion (HRE) Dataset
收藏资源简介:
Omni-Emotion数据集由阿里巴巴集团同义实验室和中山大学联合创建,包含两个子数据集:SRE数据集和HRE数据集。SRE数据集包含24,137个视频样本,通过GPT评分筛选,确保与真实情感标签高度一致;HRE数据集包含3,500个手动验证的视频样本,进一步提升了数据质量。数据集内容涵盖多种情感场景,数据来源包括DFEW、MAFW等公开数据集。创建过程中,通过稀疏采样、面部跟踪和音频提取等技术,结合GPT-3.5进行一致性检查,确保数据的高质量。该数据集主要用于多模态情感分析,旨在解决现有模型在音频和面部微表情识别上的不足,推动情感分析在真实场景中的应用。
The Omni-Emotion dataset was jointly developed by the Synonymous Laboratory of Alibaba Group and Sun Yat-sen University, and consists of two sub-datasets: the SRE dataset and the HRE dataset. The SRE dataset contains 24,137 video samples, which were screened using GPT-based scoring to ensure high consistency with ground-truth emotion labels. The HRE dataset includes 3,500 manually verified video samples, further enhancing the overall data quality. The dataset covers a wide range of emotional scenarios, with data sources including public datasets such as DFEW and MAFW. During the dataset construction process, technologies including sparse sampling, facial tracking and audio extraction were adopted, combined with GPT-3.5 for consistency checking to guarantee high-quality data. This dataset is primarily intended for multimodal emotion analysis, aiming to address the limitations of existing models in audio and facial micro-expression recognition, and promote the application of emotion analysis in real-world scenarios.
数据集概述
数据集名称
Omni-Emotion
数据集状态
- 数据集详情页面显示“coming soon”,表明该数据集目前尚未发布或详细信息尚未公开。
数据集描述
- 由于数据集详情页面显示“coming soon”,目前无法获取该数据集的具体描述、内容、用途等信息。
数据集链接




