Fast CU Depth Decision Algorithm for HEVC Intra Coding

在HEVC 参考代码中，一个CTU块通过xcompressCU()函数进行CU递归得到最优的CU深度。

递归的过程可如下图(from:Fast CU Splitting and Pruning for Suboptimal CU Partitioning in HEVC Intra Coding)所示。图中每一个方框表示一个CU块，方框内的数字表示xcompressCU()函数的执行顺序。显而易见，如果能在做xcompressCU()函数之前，将CU的递归深度确定下，显然可以减小HEVC编码器的复杂度。

针对帧内编码器，已经有很多文献提出了提前确定CU递归深度的方法。这里介绍了"Fast CU Size Decision and Mode Decision Algorithm for HEVC Intra Coding"中Section II.A部分的具体实现。在这篇文献中，周边块的CTU depth size用来给当前块深度进行预测。具体的细节可以去查看该文献。

Void TEncCu::compressCU( TComDataCU*& rpcCU )
{// initialize CU datam_ppcBestCU[0]->initCU( rpcCU->getPic(), rpcCU->getAddr() );m_ppcTempCU[0]->initCU( rpcCU->getPic(), rpcCU->getAddr() );memset( m_preAnalyzeDepth,       0, rpcCU->getTotalNumPart() );memset( m_preAnaDepthDetermined, 0, rpcCU->getTotalNumPart() );memset( m_preAnaDepthRange     , 0, rpcCU->getTotalNumPart() );// Neighboring CTUs.TComDataCU* t_pcCULeft      = rpcCU->getCULeft();TComDataCU* t_pcCUAbove     = rpcCU->getCUAbove();TComDataCU* t_pcCUAboveLeft = rpcCU->getCUAboveLeft();TComDataCU* t_pcCUAboveRight= rpcCU->getCUAboveRight();UInt DepthLeft       = 0;       // Max Depth of LeftCTU.UInt DepthAbove      = 0;       // Max Depth of AboveCTU.UInt DepthAboveLeft  = 0;UInt DepthAboveRight = 0;UInt picWidth  = rpcCU->getSlice()->getSPS()->getPicWidthInLumaSamples();UInt picHeight = rpcCU->getSlice()->getSPS()->getPicHeightInLumaSamples();UInt uiLPelX   = rpcCU->getCUPelX();UInt uiRPelX   = uiLPelX + rpcCU->getWidth(0)  - 1;UInt uiTPelY   = rpcCU->getCUPelY();UInt uiBPelY   = uiTPelY + rpcCU->getHeight(0) - 1;UChar    tDepth;m_insidePicture= (uiRPelX<picWidth) && (uiBPelY<picHeight);// Considering Border CTUs.if ( t_pcCULeft!=NULL ) //获取左边CTU块最大的depth信息{for ( Int i=0; i<256; i++ ){tDepth    = t_pcCULeft->getDepth(i);if ( tDepth>DepthLeft ){DepthLeft = (UInt)tDepth;}}}elseDepthLeft = 2; //如果是NULL，直接赋值2(16X16)if ( t_pcCUAbove!=NULL ){for ( Int i=0; i<256; i++ ){tDepth    = t_pcCUAbove->getDepth(i);if ( tDepth>DepthAbove ){DepthAbove = (UInt)tDepth;}}}elseDepthAbove = 2;if ( t_pcCUAboveLeft!=NULL ){DepthAboveLeft = t_pcCUAboveLeft->getDepth(g_auiRasterToZscan[16*15+15]);}elseDepthAboveLeft = 2;if ( t_pcCUAboveRight!=NULL ){DepthAboveRight = t_pcCUAboveRight->getDepth(g_auiRasterToZscan[16*15]);}elseDepthAboveRight = 2;Double DepthPre = 0.3*DepthLeft+0.3*DepthAbove+0.2*DepthAboveLeft+0.2*DepthAboveRight; // 论文中Prediction Depth Typeif ( DepthPre<=0.5 ) // 依据论文中的公式给出最小的depth level和最大的depth level{memset( m_preAnaDepthDetermined, 1, 256 );memset( m_preAnaDepthRange,      2, 256 );memset( m_preAnalyzeDepth,       0, 256 );}else if ( DepthPre<=1.5 ){memset( m_preAnaDepthDetermined, 1, 256 );memset( m_preAnaDepthRange,      3, 256 );memset( m_preAnalyzeDepth,       0, 256 );}else{memset( m_preAnaDepthDetermined, 1, 256 );memset( m_preAnaDepthRange,      3, 256 );memset( m_preAnalyzeDepth,       1, 256 );}DEBUG_STRING_NEW(sDebug)xCompressCU( m_ppcBestCU[0], m_ppcTempCU[0], 0 DEBUG_STRING_PASS_INTO(sDebug) );DEBUG_STRING_OUTPUT(std::cout, sDebug)// Double Check.UInt MaxDepthSize=0;// UInt CTUPelX, CTUPelY;if ( m_insidePicture ){for ( Int i=0; i<256; i++ ){// Decisioned.tDepth    = m_ppcBestCU[0]->getDepth(i);UChar cuDepth         = m_preAnalyzeDepth[i];UChar cuPreDetermined = m_preAnaDepthDetermined[i];UChar cuRange         = m_preAnaDepthRange[i];if ( tDepth<cuDepth && tDepth>=cuDepth+cuRange ){assert(0);}}}#if ADAPTIVE_QP_SELECTIONif( m_pcEncCfg->getUseAdaptQpSelect() ){if(rpcCU->getSlice()->getSliceType()!=I_SLICE) //IIII{xLcuCollectARLStats( rpcCU);}}
#endif
}

在xcompressCU函数中加入相关条件跳转。

 // If slice start or slice end is within this cu...TComSlice * pcSlice = rpcTempCU->getPic()->getSlice(rpcTempCU->getPic()->getCurrSliceIdx());Bool bSliceStart = pcSlice->getSliceSegmentCurStartCUAddr()>rpcTempCU->getSCUAddr()&&pcSlice->getSliceSegmentCurStartCUAddr()<rpcTempCU->getSCUAddr()+rpcTempCU->getTotalNumPart();Bool bSliceEnd = (pcSlice->getSliceSegmentCurEndCUAddr()>rpcTempCU->getSCUAddr()&&pcSlice->getSliceSegmentCurEndCUAddr()<rpcTempCU->getSCUAddr()+rpcTempCU->getTotalNumPart());Bool bInsidePicture = ( uiRPelX < rpcBestCU->getSlice()->getSPS()->getPicWidthInLumaSamples() ) && ( uiBPelY < rpcBestCU->getSlice()->getSPS()->getPicHeightInLumaSamples() );// Fast CU decision Process.// When Current depth is not in the PreAnalyzedDepth Range, it just skips the PU/TU Decision process.// Added by xfHuang.Bool t_enCUSkip=false;if ( m_insidePicture ){// Split Analysis For CU32X32 And CU16X16.if ( checkCurDepthInPreAnaRange( rpcBestCU, uiDepth ) == false ) //如果当前的depth level不在预测的depth level之内，后面直接将cost赋值成最大，不进行后面的预测操作。{t_enCUSkip = true;rpcBestCU->getTotalCost() = MAX_DOUBLE/16;rpcBestCU->getTotalDistortion() = MAX_UINT>>3;rpcBestCU->getTotalBits() = MAX_UINT>>3;// avoid assert disable.if ( uiDepth==3 ){rpcBestCU->setPartitionSize ( 0, SIZE_2Nx2N );      rpcBestCU->setPredictionMode( 0, MODE_INTRA );    }}}// We need to split, so don't try these modes.if(!bSliceEnd && !bSliceStart && bInsidePicture ){if( t_enCUSkip==false ){for (Int iQP=iMinQP; iQP<=iMaxQP; iQP++){const Bool bIsLosslessMode = isAddLowestQP && (iQP == iMinQP);if (bIsLosslessMode){iQP = lowestQP;}rpcTempCU->initEstData( uiDepth, iQP, bIsLosslessMode );

其中checkCurDepthInPreAnaRange函数如下：

Bool TEncCu::checkCurDepthInPreAnaRange( TComDataCU*& pCU, UInt uidepth )
{UChar cuDepth         = m_preAnalyzeDepth[pCU->getZorderIdxInCU()];UChar cuPreDetermined = m_preAnaDepthDetermined[pCU->getZorderIdxInCU()];UChar cuRange         = m_preAnaDepthRange[pCU->getZorderIdxInCU()];assert(cuDepth+cuRange<=5);if ( /*cuPreDetermined &&*/ uidepth>=cuDepth && uidepth<cuDepth+cuRange ){return true;}else{return false;}
}

以上是一种基于周边CTU块信息来进行CU深度优化的一种方法。这个方法对于大部分来说只是不做64X64这一层depth，因此性能损失很小，平均大概在0.2%左右。时间可以节省10%左右。

[转载请注明作者和出处]

Fast CU Depth Decision Algorithm for HEVC Intra Coding相关推荐

HM 复现论文 ——《Fast Intra Mode Decision for High Efficiency Video Coding》
在大致了解了 HEVC 编码标准后,这段时间我也开始尝试在 HM-16.22 中复现一些算法.这篇博客主要用于记录进一个月来的工作情况. 复现的论文标题为 Fast Intra Mode Decisi ...
【去雾论文阅读】A Novel Fast Single Image Dehazing Algorithm Based on Artificial Multiexposure Image Fusion
题目:A Novel Fast Single Image Dehazing Algorithm Based on Artificial Multiexposure Image Fusion 作者:Zh ...
HEVC intra MPM
编码帧内预测模式大量增加可选择的预测模式可以提供更高的编码效率,同时要求编码预测模式有更加高效的方法降低更多模式带来的负担.与H.264采用一个最可能模式不同的是,对于亮度分量,三个最可能模式用于预 ...
HEVC intra和inter码率控制的不同
在compressGOP()函数中 compressGOP() {...for (Int iGOPid=0;iGOPid<m_iGopSize;iGOPid++){m_pcSliceEncode ...
【FastDepth】《FastDepth：Fast Monocular Depth Estimation on Embedded Systems》
ICRA-2019 文章目录 1 Background and Motivation 2 Related Work 3 Advantages / Contributions 4 Method 5 Ex ...
视觉（5）A Fast Area-Based Stereo Matching Algorithm
L. Di Stefano, M. Marchionni, S. Mattoccia, G. Neri 1. DEIS-ARCES, University of Bologna 本文主要介绍了基于SA ...
南华大学计算机学院龚向坚,李跃-计算机科学与技术学院
李跃,男,1988年出生,教授(校聘),硕士研究生导师,湖南大学计算机科学与技术专业博士,新加坡南洋理工大学访问学生.主要研究方向为人工智能(计算机视觉).视频编码及3D点云压缩.在国内外期刊发表论文 ...
Deep Learning-Based Video Coding: A Review and A Case Study
郑重声明:原文参见标题,如有侵权,请联系作者,将会撤销发布! 1.Abstract: 本文主要介绍的是2015年以来关于深度图像/视频编码的代表性工作,主要可以分为两类:深度编码方案以及基于传统编码方 ...
南京大学计算机学院宋教授,宋云教授
[1]. 宋云,李雪玉,沈燕飞,杨高波.基于非局部相似块低秩的压缩感知图像重建算法.电子学报, 45(3):695-703, 2017. (中文权威,EI) [2]. 宋云,沈燕飞,龙际珍,朱珍民.基 ...

Fast CU Depth Decision Algorithm for HEVC Intra Coding

Fast CU Depth Decision Algorithm for HEVC Intra Coding相关推荐

最新文章

热门文章