large-scale TV dataset

Stvd-pvcd数据集处理计算机视觉中部分视频拷贝检测方法的性能评估。它是根据具有电视捕获 [1,2] 的协议设计的，可确保更深的可扩展性，强大的接地性和对降级的控制，以进行精细的性能评估。它是该任务上最大的公共数据集，具有近83k视频，总持续时间为10,660小时。
数据集由一个参考集，六个测试集a到F (呈现不同的类别和退化水平) 和基本事实组成。它作为不同的文件提供，包含:
带有ids的简短参考视频，
用于测试/培训的长正面/负面视频，
具有参考id，时间戳和持续时间的基本事实。

The Stvd-pvcd dataset is dedicated to performance evaluation of partial video copy detection methods in computer vision. It was designed following the protocol with TV capture [1,2], ensuring greater scalability, strong grounding, and control over degradations to enable fine-grained performance evaluation. It is the largest public dataset for this task, containing nearly 83,000 videos with a total duration of 10,660 hours.
The dataset consists of one reference set, six test sets A to F (covering different categories and degradation levels), and ground truth annotations. It is provided as distinct files, including:
short reference videos with unique IDs,
long positive/negative videos for testing/training,
and ground truth annotations with reference IDs, timestamps, and durations.

搜集汇总

背景与挑战

背景概述

该数据集专为视频拷贝检测性能评估设计，基于电视捕获协议构建，具备高可扩展性和对退化的精细控制。作为该任务最大的公共数据集，它包含约83k视频、总时长10,660小时，并提供参考集、多个测试集及基本事实文件。