JavisDiT/JavisBench
收藏Hugging Face2025-09-29 更新2025-11-01 收录
下载链接：
https://hf-mirror.com/datasets/JavisDiT/JavisBench
下载链接
链接失效反馈官方服务：
资源简介：
JavisBench是一个全面的基准测试，用于评估文本到音频-视频生成模型。它涵盖了生成质量、语义对齐和时序同步的多个方面，以在受控和现实世界的场景中进行全面评估。数据集由现有基准测试的测试数据和2024年6月至12月期间收集的YouTube视频组成，共包含10,140个带注释字幕和各种属性的视频音频样本。JavisBench还包含一个较小的版本JavisBench-mini，包含1,000个随机采样的样本。数据集提供了预提取的音频视频特征，用于FVD/KVD/FAD评估。
JavisBench is a comprehensive benchmark for evaluating text-to-audio-video generation models across various aspects such as generation quality, semantic alignment, and temporal synchrony. It is composed of test data from existing benchmarks and newly collected YouTube videos, with a total of 10,140 audio-video samples. The dataset also includes a smaller-scale version called JavisBench-mini with 1,000 samples. JavisBench provides pre-extracted audio-video features for FVD/KVD/FAD evaluation.
提供机构：
JavisDiT
