Long Video Dataset (3X)

我们从互联网上随机选择了三个长度超过 1.5K 帧且主要对象不断出现的视频。每个视频有 20 个均匀采样的帧，手动注释以进行评估。每个视频都经过来回播放，生成了三倍长的视频。

We randomly selected three videos from the Internet that are over 1.5K frames in length and have continuously appearing main objects. Twenty frames were uniformly sampled from each video, and these frames were manually annotated for evaluation. Each video was played back and forth to generate a video with three times the original duration.

搜集汇总

背景与挑战

背景概述

Long Video Dataset (3X) 是一个由路易斯安那州立大学和Northeastern University于2020年发布的视频数据集。该数据集包含三个从互联网选取的长度超过1.5K帧的视频，每个视频均匀采样20帧进行人工标注，并通过来回播放处理生成了三倍时长的视频序列。