LIMinghan/FiVE-Fine-Grained-Video-Editing-Benchmark
收藏Hugging Face2025-03-31 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/LIMinghan/FiVE-Fine-Grained-Video-Editing-Benchmark
下载链接
链接失效反馈官方服务：
资源简介：
FiVE Benchmark是一个包含100个视频的细粒度视频编辑评估基准数据集，旨在评估扩散和修正流模型。该数据集包括74个从DAVIS数据集抽取的真实世界视频和26个使用Wan2.1文本到视频模型生成的合成视频，提供了多样化的编辑挑战。数据集支持六个细粒度编辑任务，并提供了结构化字幕、对象变形记录以及用于模型评估的各种元素。
FiVE Benchmark is a fine-grained video editing evaluation benchmark containing 100 videos, designed for evaluating diffusion and rectified flow models. The dataset includes 74 real-world videos sampled from the DAVIS dataset and 26 highly realistic synthetic videos generated using the Wan2.1 text-to-video model, offering diverse editing challenges. It supports six fine-grained editing tasks and provides structured captions, object deformation records, and various elements for model evaluation.
提供机构：
LIMinghan
