Threshold impact for CNN partitioning algorithm in the Versatile Video Coding standard
Bouthaina Abdallah, Fatma Belghith, Mohamed Ali Ben Ayed, Nouri Masmoudi · 2023
Versatile Video Coding (VVC) standard was published in July 2020 by the Joint Video Experts Team (JVET). The VVC coding scheme has been enhanced by introducing new coding technologies compared to HEVC codec scheme. The added tools improved the Rate-Distortion (RD) performance while increasing the coding complexity. The Quadtree with Nested Multi-Type Tree (QTMT) partitioning module is one of the most complex improvement. Therefore, we suggested a low complexity QTMT algorithm based on the Convolution Neural Network (CNN). The proposed method adopted bi-threshold values in order to determine the partitioning decision according to the predicted probabilities. In this paper, a study of the bi-threshold values impact on both coding efficiency and complexity reduction is presented using the VVC reference software VTM3.0 in order to develop a method that may automatically select the appropriate bi-threshold value.