Fast VVC Inter Coding Algorithms Based On CU Depth And Mode Prediction

The rapid and continuous growth of video traffic on the Internet and the demand for higher resolution,higher video quality and immersive video are posing higher demands of transmission capabilities of digital networks.In order to alleviate the increasingly tense network bandwidth dilemma,new generations of video coding standards are consistently introduced.The Versatile Video Coding(VVC)standard released on July2020,offered nearly one time higher compression rate than the previous standard High Efficiency Video Coding(HEVC),but the coding complexity is several times higher that of HEVC,and the excessive complexity seriously hinders the promotion and application of VVC.Hence,studying fast coding algorithms for VVC has great practical significance.In order to reduce the coding complexity of VVC,we propose a novel fast VVC inter coding algorithm based on multi-task learning.The algorithm constructs a multi-task learning model to simultaneously predict all coding units(CUs)in the coding tree unit(CTU),and uses the prediction result to skip some unnecessary CU encoding processes.Hence,it boosts the speed of CU’s depth decision and CU’s mode decision process.In addition,we also propose a fast two-stage inter mode decision algorithm.In the first stage of the algorithm,the spatial correlation of the mode and the correlation between all-zero blocks and the Merge＿SKIP mode are used to quickly select the Merge＿SKIP mode for the current CU and terminate the subsequent encoding process for the current CU in advance.In the second stage of the algorithm,a binary linear regression model is constructed to predict the rate-distortion cost of inter bi-directional predictive coding.The prediction results of this model are used to early skip CU inter bi-directional predictive coding.The above algorithms are implemented on VVenc 0.3 platform,an open source implementation of VVC.Experiments show that the above algorithms can effectively reduce the coding complexity while maintaining the video quality.Among them,the fast VVC inter coding algorithm based on multi-task learning achieves averagely 10.38% encoding time reduction with 1.73% BDBR increase while the fast two-stage inter mode decision algorithm achieves 12.08% encoding time reduction with 2.37% BDBR increase.