eduzhai > Applied Sciences > Engineering >

DeepQTMT A Deep Learning Approach for Fast QTMT-based CU Partition of Intra-mode VVC

  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: Versatile Video Coding (VVC), as the latest standard, significantly improvesthe coding efficiency over its ancestor standard High Efficiency Video Coding(HEVC), but at the expense of sharply increased complexity. In VVC, thequad-tree plus multi-type tree (QTMT) structure of coding unit (CU) partitionaccounts for over 97 of the encoding time, due to the brute-force search forrecursive rate-distortion (RD) optimization. Instead of the brute-force QTMTsearch, this paper proposes a deep learning approach to predict the QTMT-basedCU partition, for drastically accelerating the encoding process of intra-modeVVC. First, we establish a large-scale database containing sufficient CUpartition patterns with diverse video content, which can facilitate thedata-driven VVC complexity reduction. Next, we propose a multi-stage exit CNN(MSE-CNN) model with an early-exit mechanism to determine the CU partition, inaccord with the flexible QTMT structure at multiple stages. Then, we design anadaptive loss function for training the MSE-CNN model, synthesizing both theuncertain number of split modes and the target on minimized RD cost. Finally, amulti-threshold decision scheme is developed, achieving desirable trade-offbetween complexity and RD performance. Experimental results demonstrate thatour approach can reduce the encoding time of VVC by 44.65 -66.88 with thenegligible Bjøntegaard delta bit-rate (BD-BR) of 1.322 -3.188 , whichsignificantly outperforms other state-of-the-art approaches.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×