Large-scale Generated Video Quality assessment (LGVQ)
收藏资源简介:
LGVQ数据集是由多个研究机构合作构建的大规模生成视频质量评估数据集,包含2808个AI生成的视频,这些视频由6种主流文本到视频生成算法基于468个文本提示生成。数据集的创建旨在从主观和客观角度系统地研究AI生成视频的质量评估问题。数据集的创建过程包括选择文本提示、生成视频、以及邀请54名参与者对视频的空间质量、时间质量和文本到视频的对齐质量进行评分。LGVQ数据集主要应用于视频生成技术的质量评估,旨在解决AI生成视频的质量评估难题,特别是在捕捉复杂失真和语义层面质量特征方面。
The LGVQ dataset is a large-scale generative video quality assessment dataset collaboratively constructed by multiple research institutions. It contains 2808 AI-generated videos, which were created based on 468 text prompts using six mainstream text-to-video generation algorithms. The dataset was developed to systematically study the quality assessment of AI-generated videos from both subjective and objective perspectives. The construction process of the dataset includes selecting text prompts, generating videos, and inviting 54 participants to score the videos on spatial quality, temporal quality, and text-to-video alignment quality. The LGVQ dataset is mainly applied to the quality assessment of video generation technologies, aiming to solve the challenges in quality assessment of AI-generated videos, particularly in capturing complex distortions and semantic-level quality features.

- 1Benchmarking AIGC Video Quality Assessment: A Dataset and Unified Model研究机构未明确提及 · 2024年



