Efficient Algorithm and Hardware Architecture for Rate Estimation in Mode Decision of AVS3
Yunyao Yan,Guoqing Xiang,Xiaodong Xie,Huizhu Jia,Yuan Li,Peng Zhang,Jie Chen
DOI: https://doi.org/10.1109/icme52920.2022.9858932
2022-01-01
Abstract:Towards enabling advanced video coding for emerging ap-plications, the AVS3 standard has been developed recently, achieving twice the coding efficiency of the AVS2 stan-dard through complex coding tools including advanced rate-distortion optimization (RDO) to select the best mode. The bit-rates are produced with the Advanced-Entropy-Coding (AEC) in the RDO process of AVS3. However, AEC dom-inates the time complexity of RDO and among the steps, con-text updating and interval subdivision are performed recur-sively, which is not conducive to real-time application, espe-cially for the hardware implementation. Thus this paper pro-poses an adaptive rate estimation algorithm with a piece-wise linear function that is very friendly to hardware implemen-tation to accelerate the rate estimation process in the RDO for AVS3 practical applications. The proposed architecture can meet the requirement of 4K@120fps ultra-high-definition videos at 200 MHz, whereas the BD-Rate increases only by 0.67% under the All-Intra (AI) configuration.