Aliyun publishes end-to-side real-time super-scoring tool: mobile video resolution can be doubled, up to 1440p

CTOnews.com Sept. 29, Aliyun Machine Learning PAI team recently released an one-button end-to-side super-scoring tool, which can double the mobile video resolution up to 1440p while keeping the device and network bandwidth unchanged.

CTOnews.com learned that super-score, or "super-resolution", refers to improving the definition of low-resolution pictures or videos through deep learning to achieve higher resolution, thereby achieving better viewing experience and recognition accuracy. By learning the known high-resolution resources through a special convolution neural network (CNN), the model of hyper-processing of low-resolution images and video resources can be realized.

This one-button-to-side hyper-scoring tool with PAI-AISR model released by Aliyun machine learning PAI team can accelerate optimization such as quantitative pruning of the model, as well as heterogeneous computing and memory I / O optimization for end-side multi-devices (CPU / GPU / NPU).

According to reports, this super-scoring tool can enhance the video resolution up to 1440p on the mobile side, and enhance the aliasing and bottom noise suppression and color enhancement at the same time, supporting cross-platform use and improving the viewing experience. Among them, the PAI-AISR algorithm only needs 50KB at least. At present, this technology has been applied in many App, such as Youku, quark, UC browser and so on.