Prnet: A Progressive Regression Network for No-Reference User-Generated-Content (UGC) Video Quality Assessment
Yang Yang, Bo Jiang, Kailin Wu · 2023
Objectively perceptual quality assessment of UGC videos is a challenge due to many reasons. First, the sources of UGC-videos are not available, which makes the appropriate technique is the no-reference video quality assessment (NR-VQA). Second, subjective mean option scores (MOS) of all the UGC-datasets are not uniformly distributed. Third, authentic degradations occurred in the videos are not unique, therefore, not predicable. For example, an over- or underexposure video, brightness and contrast static information is critical for evaluation. Only employing verified priori statistic knowledge or generalized learning knowledge may not cover all possible distortions. To solve these problems, we introduce a framework—Progressive Regression Network (PRNet) in this paper. For the MOS problem, a PR module is proposed, which utilizes the coarse-to-fine strategy during the training process. It can turn sparse subjective human rating scores into integers with denser samples to address the problem. For the unpredictable distortions, a wide and deep feature extraction module is developed, which employs both low-level features generated from natural scene statistics (NSS) and high-level semantic features extracted by deep neural networks, to fuse memorizing priori knowledge and generalizing learning features. Experiments demonstrate that our PRNet achieves top performance in three main datasets.