ResNet-Based Fast CU Partition Decision Algorithm for VVC

H.266/VVC introduces new technologies based on previous video coding standards, including the QTMT partition structure and support for 65 angular intra-prediction modes. The introduced significantly improve efficiency, but also lead to a great increase in complexity. This paper proposes ResNet-based fast CU decision algorithm reduce complexity of VVC. First, we count analyze proportion split modes explore appropriate CNN models. Then, models are designed predict mode, first convolutional layer combines symmetric asymmetric kernels extract features efficiently. We introduce RD cost into loss function prediction accuracy Finally, two threshold schemes used achieve compromise between performance. experimental results show that “Fast” scheme saves 55.12% encoding time with 1.83% BDBR “Moderate” 47.03% only 1.27% compared VTM7.0.