eduzhai > Applied Sciences > Engineering >

End-to-End Rate-Distortion Optimization for Bi-Directional Learned Video Compression

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Conventional video compression methods employ a linear transform and blockmotion model, and the steps of motion estimation, mode and quantizationparameter selection, and entropy coding are optimized individually due tocombinatorial nature of the end-to-end optimization problem. Learned videocompression allows end-to-end rate-distortion optimized training of allnonlinear modules, quantization parameter and entropy model simultaneously.While previous work on learned video compression considered training asequential video codec based on end-to-end optimization of cost averaged overpairs of successive frames, it is well-known in conventional video compressionthat hierarchical, bi-directional coding outperforms sequential compression. Inthis paper, we propose for the first time end-to-end optimization of ahierarchical, bi-directional motion compensated learned codec by accumulatingcost function over fixed-size groups of pictures (GOP). Experimental resultsshow that the rate-distortion performance of our proposed learnedbi-directional { it GOP coder} outperforms the state-of-the-art end-to-endoptimized learned sequential compression as expected.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×