Real-Time Video Super-Resolution with Spatio-Temporal Modeling and Redundancy-Aware Inference

• DOI: 10.3390/s23187880
• Publication Date: 2023-09-15T08:06:13Z
• Authors: Wenhao Wang; Zhenbing Liu; Haoxiang Lu; Rushi Lan; Zhaoyuan Zhang
• Keywords: Leverage (statistics); Temporal resolution; Frame rate

Video super-resolution aims to generate high-resolution frames from low-resolution counterparts. It can be regarded as a specialized application of image super-resolution, serving various purposes, such video display and surveillance. This paper proposes novel method for real-time super-resolution. effectively exploits spatial information by utilizing the capabilities an model leverages temporal inherent in videos. Specifically, incorporates pre-trained network its foundational framework, allowing it leverage existing expertise A fast aggregation module is presented further aggregate cues across frames. By using deformable convolution align features neighboring frames, this takes advantage inter-frame dependency. In addition, employs hierarchical offset feature extraction channel attention-based fusion. redundancy-aware inference algorithm developed reduce computational redundancy reusing intermediate features, achieving inferring speed. Extensive experiments on several benchmarks demonstrate that proposed reconstruct satisfactory results with strong quantitative performance visual qualities. The ability makes suitable real-world deployment.