遇见数据集

cminst/FLASH-Dataset

收藏
Hugging Face2025-08-15 更新2025-11-30 收录
官方服务:

资源简介:

FLASH数据集是一个用于评估模型在给定用户提示下识别视频中的关键动作帧(峰值帧)能力的开源视频评估数据集。它包含视频文件、行ID、动作评分以及关键帧的多个特征,如积累、标题、下降、峰值结束和峰值开始。该数据集的构建基于ActivityNet Captions数据集,并提供了一个比标准时序动作定位更具挑战性的任务。

FLASH is an open-source video evaluation dataset designed to assess a models capability to identify key action frames (peak frames) in a video given a users prompt. It includes video files, row IDs, action scores, and multiple features of key frames such as build_up, caption, drop_off, peak_end, and peak_start. This dataset is built upon the ActivityNet Captions dataset and provides a more challenging task than standard temporal action localization.

提供机构:
cminst
二维码
社区交流群
二维码
科研交流群
商业服务