A Machine Learning-Based Solution to Accelerate the Intra Mode Decision for the VVC Standard
Adson Duarte, A. C. A. Oliveira, Bruno Zatt, Guilherme Corrêa, Daniel Palomino · 2023
The VVC video coding standard achieves high compression rates due to innovative tools that were introduced mainly in the intra prediction. However, the high computational effort associated with the intra mode decision poses a challenge for real-time video coding applications. In this paper, we propose a machine learning-based solution to accelerate the intra mode decision of VVC. The intra modes are organized in three classes (Planar/DC, Angular and MIP) and a Decision Tree model is developed to predict the class of modes more likely to be chosen, avoiding the evaluation of the classes of modes with less chance to be the optimal. As a result, the proposed solution can reduce the total encoding time in 15.67% on average with only 0.80% of coding efficiency loss. When compared with related works, our solution presents good results.