[x265] [PATCH 2 of 2] rc: restrict low tolerance for bitrate fluctuations to strict-cbr option

[Deepthi Nandakumar]

On Tue, Dec 30, 2014 at 2:51 PM, <aarthi at multicorewareinc.com> wrote:

> # HG changeset patch
> # User Aarthi Thirumalai
> # Date 1419064540 -19800
> #      Sat Dec 20 14:05:40 2014 +0530
> # Node ID fa8ab4bd8851613983e152e682c031fd4b0e83e8
> # Parent  98dfd6e0c72122e8613208d5927a4c81da6a8e7a
> rc: restrict low tolerance for bitrate fluctuations to strict-cbr option
>
> diff -r 98dfd6e0c721 -r fa8ab4bd8851 source/encoder/ratecontrol.cpp
> --- a/source/encoder/ratecontrol.cpp    Tue Dec 23 16:32:30 2014 +0530
> +++ b/source/encoder/ratecontrol.cpp    Sat Dec 20 14:05:40 2014 +0530
> @@ -1759,7 +1759,7 @@
>                  for (int j = 0; bufferFillCur >= 0; j++)
>                  {
>                      int type = curFrame->m_lowres.plannedType[j];
> -                    if (type == X265_TYPE_AUTO)
> +                    if (type == X265_TYPE_AUTO || totalDuration >= 1.0)
>                          break;
>                      totalDuration += m_frameDuration;
>                      double wantedFrameSize = m_vbvMaxRate *
> m_frameDuration;
> @@ -1773,17 +1773,17 @@
>                      bufferFillCur -= curBits;
>                  }
>
> -                /* Try to get the buffer no more than 80% filled, but
> don't set an impossible goal. */
> -                double tol = isIFramePresent ? 1 / totalDuration :
> totalDuration < 0.5 ? 2 : 1;
> -                targetFill = X265_MIN(m_bufferFill + totalDuration *
> m_vbvMaxRate * 0.5 , m_bufferSize * (1 - 0.8 * totalDuration * tol));
> +                /* Try to get the buffer at least 50% filled, but don't
> set an impossible goal. */
> +                double tol = (isIFramePresent ? 1 / totalDuration :
> totalDuration < 0.5 ? 2 : 1 );
> +                targetFill = X265_MIN(m_bufferFill + totalDuration *
> m_vbvMaxRate * 0.5 , m_bufferSize * (1 - 0.5 * totalDuration * tol));
>                  if (bufferFillCur < targetFill)
>                  {
>                      q *= 1.01;
>                      loopTerminate |= 1;
>                      continue;
>                  }
> -                /* Try to get the buffer atleast 50% filled, but don't
> set an impossible goal. */
> -                targetFill = Clip3(m_bufferSize - (m_bufferSize *
> totalDuration * 0.5), m_bufferSize, m_bufferFill - totalDuration *
> m_vbvMaxRate * 0.5);
> +                /* Try to get the buffer not more than 80% filled, but
> don't set an impossible goal. */
> +                targetFill = Clip3(m_bufferSize * (1 - totalDuration *
> tol * 0.2), m_bufferSize, m_bufferFill - totalDuration * m_vbvMaxRate *
> 0.5);
>                  if (m_isCbr && bufferFillCur > targetFill)
>                  {
>                      q /= 1.01;
> @@ -1976,7 +1976,8 @@
>      if (row < sps.numCuInHeight - 1)
>      {
>          /* More threads means we have to be more cautious in letting
> ratecontrol use up extra bits. */
> -        double rcTol = (bufferLeftPlanned * 0.2) /
> m_param->frameNumThreads * m_param->rc.rateTolerance;
> +        double tol = m_param->rc.bStrictCbr ? 0.2 : 1;
> +        double rcTol = (bufferLeftPlanned * tol) /
> m_param->frameNumThreads * m_param->rc.rateTolerance;
>          int32_t encodedBitsSoFar = 0;
>          double accFrameBits = predictRowsSizeSum(curFrame, rce, qpVbv,
> encodedBitsSoFar);
>
> How does bStrictCbr interact with rateTolerance? Is it better to wrap
these up into a single param->cbrRateTolerance which accepts a float
tolerance value, so users can choose which point on the slope (quality vs
bitrate flatness) they prefer?


> @@ -1994,7 +1995,7 @@
>
>          while (qpVbv < qpMax
>                 && ((accFrameBits > rce->frameSizePlanned + rcTol) ||
> -                   (rce->bufferFill - accFrameBits < bufferLeftPlanned *
> 0.2) ||
> +                   (rce->bufferFill - accFrameBits < bufferLeftPlanned *
> 0.5) ||
>                     (accFrameBits > rce->frameSizePlanned && qpVbv <
> rce->qpNoVbv)))
>          {
>              qpVbv += stepSize;
> _______________________________________________
> x265-devel mailing list
> x265-devel at videolan.org
> https://mailman.videolan.org/listinfo/x265-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20141231/8035c94b/attachment-0001.html>