CDVS: Compressed Domain On Device Memory Efficient 8K Video SlowMo

Jing Li, Chengyu Wang, Hamid R. Sheikh, SeokJun Lee; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025, pp. 1972-1978

Video slow-motion (slow-mo) is being increasingly used in a wide variety of applications ranging from entertainment, sports analysis, medical imaging, surveillance and forensics. As mobile device captured video resolution and quality increases, there is growing need for mobile video processing systems. However, in order to be effective, mobile based video slow-mo systems require memory efficient and computationally efficient algorithms. In this article, we propose a general system for a highly memory and computationally efficient Compressed Domain on device 8K Video Slow-mo (CDVS) algorithm. The system is based on an encoder decoder structured image compression network, followed by a series of efficient processing blocks within the compressed domain. The algorithm is validated on the real captured 8K videos from a mobile phone. In addition, we deploy the algorithms on mobile devices. Results indicate that the proposed algorithm can be implemented on a commodity mobile device and robustly achieves high video qualities under different conditions.