Real-Time Video Super-Resolution on Smartphones with Deep Learning, Mobile AI 2021 Challenge: Report
Andrey Ignatov,Andres Romero,Heewon Kim,Radu Timofte,Chiu Man Ho,Zibo Meng,Kyoung Mu Lee,Yuxiang Chen,Yutong Wang,Zeyu Long,Chenhao Wang,Yifei Chen,Boshen Xu,Shuhang Gu,Lixin Duan,Wen Li,Wang Bofei,Zhang Diankai,Zheng Chengjian,Liu Shaoli,Gao Si,Zhang Xiaofeng,Lu Kaidi,Xu Tianyu,Zheng Hui,Xinbo Gao,Xiumei Wang,Jiaming Guo,Xueyi Zhou,Hao Jia,Youliang Yan
DOI: https://doi.org/10.1109/cvprw53098.2021.00287
2021-06-01
Abstract:Video super-resolution has recently become one of the most important mobile-related problems due to the rise of video communication and streaming services. While many solutions have been proposed for this task, the majority of them are too computationally expensive to run on portable devices with limited hardware resources. To address this problem, we introduce the first Mobile AI challenge, where the target is to develop an end-to-end deep learning-based video super-resolution solutions that can achieve a real-time performance on mobile GPUs. The participants were provided with the REDS dataset and trained their models to do an efficient 4X video upscaling. The runtime of all models was evaluated on the OPPO Find X2 smartphone with the Snapdragon 865 SoC capable of accelerating floating-point networks on its Adreno GPU. The proposed solutions are fully compatible with any mobile GPU and can upscale videos to HD resolution at up to 80 FPS while demonstrating high fidelity results. A detailed description of all models developed in the challenge is provided in this paper.