Technical Papers: Watch presentations by the authors and read their technical papers on machine learning, including an introduction to Supernova, a deep learning-based image/video quality enhancement platform and the application of machine learning in video processing. 

Technical papers: Machine Learning

In this session we look at two state of the art machine learning based video enhancement systems. The paper from SK Telekom reports on a tool that performs up-conversion, frame interpolation and aspect ratio adjustment, common tasks for any content provider.  Whilst the paper from the BBC demonstrates a machine learnt up-conversion filter, offering a demonstrably useful improvement in modern video codec design and asks, can you trust your algorithm? Both a serious and a fascinating question when you consider the risk of bias as your algorithm goes about “improving” skin tones and “enhancing” facial features. There are also two excellent supporting papers from RheinMain University and Shanghai Jiao Tong University, which between them discuss machine learnt inverse tone mapping and gamut extension (of particular relevance in HDR) as well as up-conversion. 

Presenters: Taeyoung Na, Manager, SK Telecom 

Luka Murn, R&D Engineer, BBC 

Host: Dr. Paul Entwistle, IBC Technical Papers Committee 

Technical Papers

Read Technical Paper: Introduction to SUPERNOVA

Read Technical Paper: Towards transparent application of machine learning in video processing

Supporting Papers

Read Deep-learning-based inverse tone mapping operations for HDR image reconstruction in broadcast

Read A single convolutional neural network for joint super-resolution gamut extension, and inverse tone-mapping network