ClipShots

ClipShots 是一个大规模的镜头边界检测数据集，收集自 Youtube 和微博，涵盖 20 多个类别，包括体育、电视剧、动物等。 TRECVID 和 RAI 仅包含帧相对静止的纪录片或脱口秀节目，ClipShots 包含来自 Youtube 和微博的大量短视频。许多短视频都是自制的，挑战更多，例如手持振动和大遮挡。这些视频的类型多种多样，包括电影聚光灯、比赛花絮、手机录制的家庭视频等，每个视频的长度为1-20分钟。数据集中的渐变包括溶解、淡入淡出和滑入滑出。ClipShots is a large-scale shot boundary detection dataset collected from YouTube and Weibo, covering over 20 categories including sports, TV series, animals, and more. Unlike TRECVID and RAI, which only contain documentary or talk show programs with relatively static frames, ClipShots includes a vast number of short videos from these two platforms. Many of these short videos are user-generated, which poses greater challenges such as handheld camera shake and heavy occlusions. The types of these videos are diverse, including movie spotlights, game highlights, home videos recorded on mobile phones, etc., with each video ranging from 1 to 20 minutes in duration. The transitions in the dataset include dissolve, fade in/fade out, and slide in/slide out.

搜集汇总

背景与挑战

背景概述

ClipShots是一个大规模的视频镜头边界检测数据集，收集自Youtube和微博，涵盖20多个类别，包含多样化的视频类型和复杂场景，如手持振动和大遮挡。数据集大小为45.0GB，包含20.5k个视频，每个视频长度为1-20分钟，适用于视频场景过渡检测和视频截取研究。