MACHINE LEARNING BASED AV2 ENCODER/DECODER EFFORTS
Abstract
This paper provides a summary of the machine learning components incorporated into the AV2 video codec developed by the Alliance for Open Media(AOM). Building upon AV1, the
integration of ML-based decoding techniques, such as datadriven intra-prediction along with encoder speed optimizations, is essential for improving coding efficiency and refining
search space pruning. This paper covers the specific design of these tools, alongside presenting results from tool-off experiments and demonstrating encoding speed improvements
under standard test conditions.
integration of ML-based decoding techniques, such as datadriven intra-prediction along with encoder speed optimizations, is essential for improving coding efficiency and refining
search space pruning. This paper covers the specific design of these tools, alongside presenting results from tool-off experiments and demonstrating encoding speed improvements
under standard test conditions.
