Shufflenetv2-Based Coding Unit Partition in Intra Coding

High Efficiency Video Coding (HEVC) is one of the most widely used coding standards because of its significant improvement in coding efficiency that has been noticed by many researchers both at home and abroad. Unlike H.264(AVC), HEVC contains a quadtree-based block partitioning method, which gives it a great coding complexity. We propose a ShuffleNet-V2-based convolutional neural network (CNN) model to accomplish fast coding unit partitioning in intra coding, and constructs a largescale dataset to train the CNN to predict the depth labels of coding unit (CU) segmentation in advance, aiming to get rid of the recursive rate distortion optimization (RDO) computational process, thus effectively reducing the coding complexity. By conducting experiments on video coding standard test sequences, the algorithm proposed in this paper outperforms most of the state-of-the-art methods in the field of fast intra-frame CU segmentation by reducing the coding time by 78.95% compared to the official HEVC test model (HM16.20) with only a drop in BDBR 4.10%. The design of this paper is more favorable to be deployed in small volume and low computational power devices for online encoding.