Machine learning compression

Research group on the applications of machine learning to compression

Seminar of Li WANG: Enhancing HEVC spatial prediction by context-based learning

Li Wang will talk about our recent work on improving HEVC spatial prediction using an encoder-decoder network.

The seminar will be at Telecom Paristech on Dec 19, at 10 am.

Title: Enhancing HEVC spatial prediction by context-based learning


In HEVC compression, despite the high number of available prediction modes, current spatial prediction approaches assume the underlying signal can be approximated by a single linear combination of a few reconstructed pixels. However, increasing further the number of prediction modes will continuously increase the computational cost, and is still ineffective when the signal to predict requires more complex representations.

Recently, deep generative models such as auto-encoders have been employed to compress images, image residuals or to predict image regions. In this work we study how learning-based approaches might be used to enhance the HEVC prediction. We proposed a context based prediction enhancement model to reduce the energy of the residuals and thus improve the coding performance. We show some preliminary results and ongoing work on the topic.

Comments are closed.