eduzhai > Applied Sciences > Engineering >

Conditional Entropy Coding for Efficient Video Compression

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 23 pages

Abstract: We propose a very simple and efficient video compression framework that onlyfocuses on modeling the conditional entropy between frames. Unlike priorlearning-based approaches, we reduce complexity by not performing any form ofexplicit transformations between frames and assume each frame is encoded withan independent state-of-the-art deep image compressor. We first show that asimple architecture modeling the entropy between the image latent codes is ascompetitive as other neural video compression works and video codecs whilebeing much faster and easier to implement. We then propose a novel internallearning extension on top of this architecture that brings an additional 10 bitrate savings without trading off decoding speed. Importantly, we show thatour approach outperforms H.265 and other deep learning baselines in MS-SSIM onhigher bitrate UVG video, and against all video codecs on lower framerates,while being thousands of times faster in decoding than deep models utilizing anautoregressive entropy model.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×