Learning Perceptual Representations for Gaming NR-VQA with Multi-Task FR Signals

Researchers developed MTL-VQA, a multi-task learning framework that automatically assesses gaming video quality without needing human-labeled training data. By using full-reference metrics as supervisory signals, the system learns perceptual representations that transfer effectively to no-reference quality assessment. The approach achieves state-of-the-art performance on gaming video datasets, handling challenges like fast motion, stylized graphics, and compression artifacts that traditional methods struggle with in gaming content.