CNN-based Parameter Decision for R-lambda Model for VVC Intra Rate Control

The R-$\lambda$ model parameters of each coding tree unit (CTU) are not updated with the coding order within the same frame, it is difficult to determine the ideal initial model parameters for each intra frame in rate control. This paper proposes a convolutional neural network (CNN)-based optimal model parameter decision for Versatile Video Coding (VVC) intra rate control. Initially, a CNN dataset is constructed using the natural image dataset and the default rate control model. A CNN model with two outputs is developed to predict the significant R-$\lambda$ model parameters $\boldsymbol{\alpha}$ and $\boldsymbol{\beta}$. Different CNN models are then trained based on the threshold of $\alpha$ in the dataset, and the model with the greatest potential for rate control efficiency improvement is chosen to be embedded in the encoder. Experimentally, the method improves rate control accuracy by 2.33 percent.