Abstract
The development of the Internet and new media technology has given rise to various forms of media to meet users’ usage needs and expectations. User-Generated-Content (UGC) videos have become a way for users to maximize their participation and express themselves. A large number of UGC videos are uploaded every day on platforms such as Tiktok, Kwai, and Station B. However, the quality of UGC videos is also mixed, with videos with poor image quality providing viewers with a lower user experience. Media service providers need to compress videos uploaded by users to reduce transmission bit rates. How to evaluate the quality of compressed UGC videos is crucial for selecting an appropriate bit rate. In this paper, we have established a UGC database called UGC-New. Concretely, 405 UGC source videos are collected from major video platforms and we encoded the original video using H.264 and H.265 to obtain 2430 distorted videos, resulting in a total of 2835 videos in the dataset. We conducted a subjective human study on this database and obtained 56700 human quality ratings recorded by 100 subjects. At the same time, we also tested a series of VQA models on the dataset for objective analysis and evaluation, including a new one, called New-VQA.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call