Deep Video Quality Assessment

Deep video quality assessment (VQA) aims to automatically evaluate video quality, mirroring human perception, often without needing a pristine reference video (no-reference VQA). Current research heavily focuses on improving the efficiency of deep learning-based VQA models, particularly for high-resolution videos, by employing techniques like novel sampling strategies (e.g., full-pixel covering, mini-cube sampling) and efficient architectures (e.g., Swin Transformers, Recurrent Memory Transformers). These advancements are crucial for enabling real-time video quality monitoring and optimization in various applications, such as video streaming and conferencing, where computational constraints are significant.