Fast CU partition algorithm based on swin-transformer for depth intra coding in 3D-HEVC

The three-dimensional High Efficiency Video Coding (3D-HEVC) standard is an extension of the High Efficiency Video Coding (HEVC) standard which is the latest three-dimensional (3D) video coding standard available. Based on the HEVC standard, 3D-HEVC adds some advanced techniques that are conducive to depth map coding at the expense of a significant increase in coding complexity. In this paper, a Swin-CNN network is proposed, which leverages the advantages of Swin Transformer in extracting global information and convolutional neural network (CNN) in extracting local information. Through Swin-CNN, the coding tree unit (CTU) partition structure in depth intra coding (DIC) can be predicted accurately. In addition, we construct a large-scale depth map dataset to train the Swin-CNN. Finally, we use the proposed algorithm to replace the search process of CTU quadtree partition in DIC. Experimental results show that the proposed algorithm can reduce the coding time by 60.9% to 67.5% without compromising the quality of the synthesised views, effectively reducing the coding complexity of 3D-HEVC.