LIVE-VQC (LIVE Video Quality Challenge (VQC) Database)

摄像技术、相机设计、压缩和处理协议、通信和带宽环境以及显示方面的巨大差异导致了各种各样的视频障碍。当前的无参考 (NR) 视频质量模型无法处理这种失真的多样性。这在一定程度上是正确的，因为可用的视频质量评估数据库包含非常有限的内容、固定的分辨率，是由少数摄像师使用少量相机设备捕获的，并且受到了适度的失真。因此，这些数据库无法充分代表真实世界的视频，这些视频包含在高度多样化的成像条件下获得的非常不同种类的内容，并且受到难以或不可能模拟的真实、复杂且经常混合的失真的影响。因此，在真实世界视频数据上测试的 NR 视频质量预测器通常表现不佳。为了推进 NR 视频质量预测，我们构建了一个大型视频质量评估数据库，其中包含 585 个独特内容的视频，这些视频由 80 个不同的用户使用 101 个不同的设备（43 个设备模型）捕获，具有各种级别的复杂、真实的失真。我们通过众包收集了大量主观视频质量分数。共有 4776 名独特的参与者参与了这项研究，产生了超过 205000 个意见分数，每个视频平均记录了 240 个人类意见。这项研究是有史以来规模最大的视频质量评估研究，涉及几个关键维度：独特内容的数量、捕获设备、失真类型和失真组合、研究参与者和记录的主观分数。

Vast discrepancies in imaging technologies, camera designs, compression and processing protocols, communication and bandwidth environments, and displays give rise to a wide range of video impairments. Current no-reference (NR) video quality models fail to handle this diversity of distortions. This is partially true, as existing video quality assessment databases contain extremely limited content, fixed resolutions, are captured by a small number of photographers using a limited set of camera equipment, and suffer from mild distortions. Consequently, these databases fail to adequately represent real-world videos, which contain highly diverse content acquired under a wide range of imaging conditions and are subjected to real, complex, and often mixed distortions that are difficult or impossible to simulate. As a result, NR video quality predictors tested on real-world video data often perform poorly. To advance NR video quality prediction research, we constructed a large-scale video quality assessment database containing 585 videos with unique content, captured by 80 distinct users using 101 different devices (43 device models) with varying levels of complex, real-world distortions. We collected a large corpus of subjective video quality scores via crowdsourcing. A total of 4,776 unique participants joined this study, generating over 205,000 opinion scores, with an average of 240 human opinions per video. This study represents the largest video quality assessment research to date, spanning several key dimensions: the number of unique content items, capture devices, distortion types and combinations, study participants, and recorded subjective scores.

搜集汇总

背景与挑战

背景概述

LIVE-VQC是一个大规模视频质量评估数据库，包含585个由不同用户和设备捕获的独特视频，这些视频具有复杂且真实的失真。通过众包收集了4776名参与者的主观评分，共产生超过205000个意见分数，旨在推进无参考视频质量预测的研究。