FDA (Facial Dynamics Annotation)
收藏资源简介:
FDA数据集是由阿里巴巴集团和南开大学联合创建的高质量视频数据集,专门用于动态面部表情描述任务。该数据集包含5,033个手动标注的高质量视频片段,涵盖了超过700,000个标注词。数据来源包括现有情感视频数据集和通过网络爬取的自收集数据,确保了数据的多样性和丰富性。数据集创建过程中,研究人员通过精心设计的提示词生成初步标注,并经过人工校正,确保标注的准确性和详细性。该数据集的应用领域主要集中在提升视频多模态大语言模型在面部表情识别和描述任务中的表现,旨在解决现有模型在面部细节编码和描述能力上的不足。
The FDA dataset is a high-quality video dataset jointly created by the Alibaba Group and Nankai University, specifically designed for the task of dynamic facial expression description. The dataset contains 5,033 manually annotated high-quality video clips, covering over 700,000 annotated words. The data sources include existing emotional video datasets and self-collected data gathered through web crawling, ensuring the diversity and richness of the data. During the dataset creation process, researchers generated preliminary annotations through carefully designed prompts and then corrected them manually to ensure the accuracy and detail of the annotations. The primary application domain of this dataset focuses on enhancing the performance of video multimodal large language models in facial expression recognition and description tasks, aiming to address the deficiencies in the encoding and descriptive capabilities of existing models in facial details.




