TDVE-DB
收藏资源简介:
TDVE-DB是一个大规模的基准数据集,用于文本驱动视频编辑的质量评估。该数据集由来自12个不同模型生成的3857个编辑视频组成,涵盖了8个编辑类别,并附有173,565个来自人类主观评分的注释,涵盖了三个关键维度:编辑视频质量、编辑对齐和结构一致性。TDVE-DB的创建是为了解决文本驱动视频编辑技术中缺乏专门的视频质量评估模型的问题。数据集的构建过程包括数据收集、视频编辑评价维度的确定、主观视频编辑质量评估和数据分析。TDVE-DB旨在为全面评估现有视频编辑模型提供坚实的基础,并促进视频编辑质量评估领域的研究。
TDVE-DB is a large-scale benchmark dataset for quality assessment of text-driven video editing. It consists of 3,857 edited videos generated by 12 distinct models, covering 8 editing categories, and is accompanied by 173,565 human subjective scoring annotations spanning three core dimensions: edited video quality, editing alignment, and structural consistency. TDVE-DB was developed to address the gap in specialized video quality assessment models for text-driven video editing technologies. The construction of this dataset encompasses data collection, the definition of evaluation dimensions for video editing, subjective assessment of video editing quality, and data analysis. TDVE-DB aims to provide a solid foundation for comprehensive evaluation of existing video editing models and promote research in the field of video editing quality assessment.




