Abstract
The Enhanced Video Coding (EVC) workgroup of the Moving Picture, Audio and Data Coding by Artificial Intelligence (MPAI) organization aims at enhancing traditional video codecs by improving or replacing traditional encoding tools with AI-based counterparts. In this work, we explore enhancing MPEG Essential Video Coding (EVC) intra prediction with a learnable predictor: we recast the problem as a hole inpainting task that we tackle via masked convolutions. Our experiments in standard test conditions show BD-rate reductions in excess of 6% over the EVC baseline profile reference with some sequences in excess of 12%. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
Type
Publication
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Gabriele Spadaro
Postdoc
