STDF-DSCS: A Lightweight Scheme for Compressed Video Quality Enhancement

Lossy compression reduces the data amount in the video by sacrificing quality, which leads to severe distortion, especially when videos are overly compressed. Con-sequently, many restoration methods have been proposed to repair compressed videos. However, they are limited by their substantial computational demand and complexity. To conserve computational resources and enhance real-time capabilities, we implement a compressed video enhancement method based on Spatio-Temporal Deformable Fusion (STDF) and subsequently design a lightweight scheme for it. By integrating depthwise separable convolution and channel shuffle techniques, we design a lightweight version of STDF called STDF-DSCS. STDF-DSCS reduces computational complexity and improves model inference speed while maintaining enhancement quality. The depthwise separable convolution reduces the model's computational complexity, while channel shuffle compensates for the shortcomings of group convolution, enhancing inter-channel communication and the network's feature learning capability. Extensive experiments demonstrate that STDF-DSCS significantly enhances computational efficiency while maintaining comparable enhancement effects to existing methods, which can be utilized in real-time video processing scenarios.