A Lightweight CNN for Low-Complexity HEVC Intra Encoder
Xiaobo Guo, Qin Wang, Jianfei Jiang · 2020
The High-Efficiency Video Coding (HEVC) standard supports the quadtree division of coding unit (CU), which brings huge computational overhead and encoding delay in actual application. To simplify the CU mode decision process, this paper proposed a fast CU mode decision algorithm based on a lightweight convolutional neural network (L-CNN). The proposed scheme was implemented on HM16.5. According to the experimental results, the BD-BR of the proposed scheme only increased by 2.49%, BD-PSNR only decreased by 0.12 with the All Intra (AI) configuration, with saving 62.29% encoding time.