A Constrained Variable Bit Rate (CVBR) Algorithm for VVenC, an Open VVC Encoder Implementation

Christian R. Helmrich, Senior Member, IEEE, Christian Bartnik, Jens Bran­denburg, Adam Wieckowski, Benjamin Bross, and Detlev Marpe, Fellow, IEEE

The above paper has been accepted for presentation at the IEEE Int. Conf. on Visual Communications and Image Processing (VCIP), Jeju, in December 2023.  This page provides supplementary information on the publication, including more detailed sequence-wise evaluation results.

Please note that there is also a preceding introductory paper on VVenC's rate control algorithm, presented at the IEEE VCIP, Munich in December 2021, on which this publication is based.  That paper and comments thereon are made available link to sub-page here.

An open-source implementation of the XPSNR algorithm is freely available on GitHub:  link to external web page XPSNR filter plug-in for FFmpeg

Contents:

  1. Corrections to the IEEE VCIP published version of the paper as of September 2023

  2. Further evaluation results incl. sequence-wise XPSNR, BitErr, and rate outlier data

  1. Corrections to the IEEE VCIP published version of the paper as of September 2023

    Minor details were corrected or clarified.  The first submitted version of the paper is available here:  link to document Paper manuscript
    In the final paper revision, a reference to this supplementary page, containing the additional information requested by the reviewers, was added.

    The following information is provided to address the reviewers' comments or requests:

    • Fig. 2: The frame-wise and GOP-wise curves for Rmax = 1.5·Rtarget are faulty. The final paper includes a correct and easier-to-read figure.

    • BitErr: The RC rate accuracy measure, adopted from [1], is defined as follows, where Ractual is the size, in bits, of the resulting bitstream:
                       | RtargetRactual |
      BitErr  =  ———————  · 100%,       excluding any container data overhead.
                                Rtarget
      Hence, RC rate matching accuracy is being quantified in percent, and the higher the value of BitErr, the greater the deviation from Rtarget.

    • The text below Tab. II should read: “the results for the sequence-wise RC ... the noncapped sequence-wise baseline RC are very similar,”

    • Due to space limitations, Tables I and II could not be extended with more rate related data.  Hence, the requested data are provided below.

  2. Further evaluation results incl. sequence-wise XPSNR, BitErr, and rate outlier data

    The following comma separated value (CSV) files contain tables with per-sequence XPSNR [2], BitErr, and maximum (i.e. peak) instantaneous rate measurements (Ri in the paper), with the latter determined GOP-wise using a sliding analysis window of length Intra period  = 1 or 2 GOPs.

    All rate values are tabulated in kilobit/s (kbps), with outliers marked by a !, all XPSNR values are listed in decibel (dB); the tabulator width is 8.

    All videos are 4:2:0 color subsampled and 10 seconds long.  The resolution, frame rate (fps), and bit depth of each video sequence is as follows:

     Class 


     Name of Sequence 


     Width 


     Height 


     fps 


     Bit Depth 


     UHD
     A

     Tango4K
     FoodMarket
     Campfire
     CatRobot
     DaylightRoad
     ParkRunning

     4096
     3840
     3840
     3840
     3840
     3840

     2160
     2160
     2160
     2160
     2160
     2160

     60
     60
     30
     60
     60
     50

       10
       10
       10
       10
       10
       10

     UHD
     HHI

     BerlinCrossroads
     ChestnutTree
     March18thSquare
     NeptuneFountain
     Oberbaum
     Quadriga
     ReichstagIntoTree
     Spree

     3840
     3840
     3840
     3840
     3840
     3840
     3840
     3840

     2160
     2160
     2160
     2160
     2160
     2160
     2160
     2160

     60
     60
     60
     60
     60
     60
     60
     60

       10
       10
       10
       10
       10
       10
       10
       10

     HD
     B

     MarketPlace
     RitualDance
     Cactus
     BasketballDrive
     BQTerrace

     1920
     1920
     1920
     1920
     1920

     1080
     1080
     1080
     1080
     1080

     60
     60
     50
     50
     60

       10
       10
        8
        8
        8

     HD
     HHI

     BerlinCrossroads
     ChestnutTree
     March18thSquare
     NeptuneFountain
     Oberbaum
     Quadriga
     ReichstagIntoTree
     Spree

     1920
     1920
     1920
     1920
     1920
     1920
     1920
     1920

     1080
     1080
     1080
     1080
     1080
     1080
     1080
     1080

     60
     60
     60
     60
     60
     60
     60
     60

       10
       10
       10
       10
       10
       10
       10
       10

     SD
     C

     BasketballDrill
     BQMall
     PartyScene
     RaceHorses

      832
      832
      832
      832

      480
      480
      480
      480

     50
     60
     50
     30

        8
        8
        8
        8

     SD
     D

     BasketballPass
     BQSquare
     BlowingBubbles
     RaceHorses

      416
      416
      416
      416

      240
      240
      240
      240

     50
     60
     50
     30

        8
        8
        8
        8

Please note that the layout or content of this web page may change.  If it does, the date in the following line will be updated.

Christian R. Helmrich, September 30, 2023



References

[1]   Z. Wang, A. Rehman, K. Zheng, J. Wang, and Z. Wang, “SSIM-Motivated Two-Pass VBR Coding for HEVC,” IEEE
       Trans. Circuits Systems for Video Technology, vol. 27, no. 10, pp. 2189 2203, Oct. 2017. Online:  link to external web page IEEE Xplore

[2]   C. R. Helmrich, S. Bosse, H. Schwarz, D. Marpe, and T. Wiegand,  “A Study of the Extended Perceptually Weighted
       Peak Signal-to-Noise Ratio (XPSNR) for Video Compression with Different Resolutions and Bit Depths,” ITU Jour-
       nal: ICT Discoveries The Future of Video and Immersive Media, vol. 3, no. 1, May 2020. Online:  link to external web page ITU Journal