CNN-LNN Based Fast CU Partitioning Decision for VVC 3D Video Depth Map Intra Coding

Currently, the coding efficacy of cutting-edge video standard H.266/VVC surpasses that 3D-HEVC (3D-High Efficiency Video Coding), but existing VVC (Versatile Coding) low-complexity algorithm is mainly optimized for 2D and cannot fully utilize characteristics depth map itself. Based on this, we propose a fast decision employing CNN (Convolutional Neural Network)-LNN (Lightweight Network) model to diminish intricacy intra in 3D video. The treats CU partitioning process as two-stage process, first adding non-local block spatial pyramid pooling model, enabling proposed skip flat regions perform adaptive prediction CUs edge regions; then, LNN used make early TT (Ternary Tree) partition need be partitioned, decisions do not partitioned by TT, so reduce some unnecessary RDO calculations. Experimental results illustrate achieves notable reduction encoding time amounting 43.23% average, with negligible impact increase BDBR.