YT-BB (YouTube-BoundingBoxes)

YouTube-BoundingBoxes (YT-BB) 是一个大规模的视频 URL 数据集，具有密集采样的对象边界框注释。该数据集由大约 380,000 个大约 19 秒长的视频片段组成，这些片段被自动选择以在自然环境中呈现对象，无需编辑或后处理，其记录质量通常类似于手持手机摄像头。这些对象代表 MS COCO 标签集的一个子集。所有视频片段均以每秒 1 帧的速度使用高精度分类标签和边界框进行人工注释。

YouTube-BoundingBoxes (YT-BB) is a large-scale video URL dataset with densely sampled object bounding box annotations. The dataset consists of approximately 380,000 video clips each about 19 seconds long. These clips are automatically selected to feature objects in natural environments, without any editing or post-processing, and their recording quality is typically comparable to that of handheld mobile phone cameras. The objects covered belong to a subset of the MS COCO label set. All video clips are manually annotated at a rate of 1 frame per second with high-precision classification labels and bounding boxes.

搜集汇总

背景与挑战

背景概述

YT-BB (YouTube-BoundingBoxes) 是一个大规模视频数据集，包含约38万个约19秒长的视频片段，这些片段在自然环境中自动选择以呈现对象，记录质量类似手持手机摄像头。数据集以每秒1帧的速度进行人工注释，提供高精度的分类标签和对象边界框，对象标签基于MS COCO标签集的子集，适用于视频对象检测和跟踪任务。