From cddbce8bca419a48dcf60be86f5f14b69678d4ab Mon Sep 17 00:00:00 2001
From: Renjie Chang <renjiechang@tencent.com>
Date: Thu, 2 Mar 2023 13:19:16 +0000
Subject: [PATCH] SR configuration for rate matching
---
README.md | 5 +++++
cfg/nn-based/nnsr_classA1.cfg | 2 ++
cfg/nn-based/nnsr_classA2.cfg | 2 ++
source/App/EncoderApp/EncApp.cpp | 2 ++
source/App/EncoderApp/EncAppCfg.cpp | 4 +++-
source/App/EncoderApp/EncAppCfg.h | 2 ++
source/Lib/CommonLib/CommonDef.h | 5 +++++
source/Lib/EncoderLib/EncCfg.h | 8 +++++++-
source/Lib/EncoderLib/EncLib.cpp | 26 +++++++++++++++++++++++---
9 files changed, 51 insertions(+), 5 deletions(-)
create mode 100644 cfg/nn-based/nnsr_classA1.cfg
create mode 100644 cfg/nn-based/nnsr_classA2.cfg
diff --git a/README.md b/README.md
index b77b5b8a47..3d54053ab7 100644
--- a/README.md
+++ b/README.md
@@ -419,10 +419,15 @@ please add the following argument when running the VTM-11-NNVC encoder/decoder e
where `path_to_directory_models_intra` is the path to the directory "models/intra" relatively to the directory from which the
VTM-11-NNVC encoder/decoder executable is run.
+
NN-based super resolution
------------------------------------------------------------------------
To activate NN-based super resolution, use --NnsrOption=1. The default model path is set as "./models/super_resolution/".
+For rate matchinng, use following config file when testing class A1 or A2.
+[cfg/nn-based/nnsr_classA1.cfg](cfg/nn-based/nnsr_classA1.cfg)
+[cfg/nn-based/nnsr_classA2.cfg](cfg/nn-based/nnsr_classA2.cfg)
+
Content-adaptive post-filter
------------------------------------------------------------------------
diff --git a/cfg/nn-based/nnsr_classA1.cfg b/cfg/nn-based/nnsr_classA1.cfg
new file mode 100644
index 0000000000..593f428836
--- /dev/null
+++ b/cfg/nn-based/nnsr_classA1.cfg
@@ -0,0 +1,2 @@
+NnsrQpOffset : 5
+NnsrQpOffsetOverrideByQp : 22 5 27 5 32 5 37 5 42 5
diff --git a/cfg/nn-based/nnsr_classA2.cfg b/cfg/nn-based/nnsr_classA2.cfg
new file mode 100644
index 0000000000..36fb542cdf
--- /dev/null
+++ b/cfg/nn-based/nnsr_classA2.cfg
@@ -0,0 +1,2 @@
+NnsrQpOffset : 5
+NnsrQpOffsetOverrideByQp : 22 10 27 6 32 6 37 6 42 6
diff --git a/source/App/EncoderApp/EncApp.cpp b/source/App/EncoderApp/EncApp.cpp
index 09b55af765..6ed57bdb3f 100644
--- a/source/App/EncoderApp/EncApp.cpp
+++ b/source/App/EncoderApp/EncApp.cpp
@@ -263,6 +263,8 @@ void EncApp::xInitLibCfg()
#if JVET_AC0196_NNSR
m_cEncLib.setUseNnsr (m_nnsrOption);
+ m_cEncLib.setNnsrQpOffset (m_nnsrQpOffset);
+ m_cEncLib.setNnsrQpOffsetOverridebyQp (m_sNnsrQpOffsetOverrideByQp);
#endif
m_cEncLib.setProfile ( m_profile);
diff --git a/source/App/EncoderApp/EncAppCfg.cpp b/source/App/EncoderApp/EncAppCfg.cpp
index 694e1c9633..a35d50009d 100644
--- a/source/App/EncoderApp/EncAppCfg.cpp
+++ b/source/App/EncoderApp/EncAppCfg.cpp
@@ -1472,6 +1472,8 @@ bool EncAppCfg::parseCfg( int argc, char* argv[] )
( "RPR", m_rprEnabledFlag, true, "Reference Sample Resolution" )
#if JVET_AC0196_NNSR
( "NnsrOption", m_nnsrOption, false, "NN-based super resolution option")
+ ( "NnsrQpOffset", m_nnsrQpOffset, 5, "Base QP offset for Rate matching")
+ ( "NnsrQpOffsetOverrideByQp", m_sNnsrQpOffsetOverrideByQp, string(""), "Override QP offset for Rate matching")
( "ScalingRatioHor", m_scalingRatioHor, 2.0, "Scaling ratio in hor direction" )
( "ScalingRatioVer", m_scalingRatioVer, 2.0, "Scaling ratio in ver direction" )
( "FractionNumFrames", m_fractionOfFrames, 1.0, "Encode a fraction of the specified in FramesToBeEncoded frames" )
@@ -1678,7 +1680,7 @@ bool EncAppCfg::parseCfg( int argc, char* argv[] )
const list<const char*>& argv_unhandled = po::scanArgv(opts, argc, (const char**) argv, err);
#if JVET_AC0196_NNSR
- m_nnsrOption = m_nnsrOption ? true : false;
+ m_nnsrOption = (m_nnsrOption && m_iSourceWidth >= NNSR_ENABLING_WIDTH && m_iSourceHeight >= NNSR_ENABLING_HEIGHT) ? true : false;
m_scalingRatioHor = m_nnsrOption ? 2.0 : 1.0;
m_scalingRatioVer = m_nnsrOption ? 2.0 : 1.0;
m_upscaledOutput = (m_nnsrOption && (m_upscaledOutput != 0)) ? 2 : 0;
diff --git a/source/App/EncoderApp/EncAppCfg.h b/source/App/EncoderApp/EncAppCfg.h
index a4c6a1c58d..eda98c8ff6 100644
--- a/source/App/EncoderApp/EncAppCfg.h
+++ b/source/App/EncoderApp/EncAppCfg.h
@@ -783,6 +783,8 @@ protected:
#if JVET_AC0196_NNSR
bool m_nnsrOption;
+ int m_nnsrQpOffset;
+ std::string m_sNnsrQpOffsetOverrideByQp;
#endif
#if JVET_AC0055_NN_POST_FILTERING
bool m_nnpf;
diff --git a/source/Lib/CommonLib/CommonDef.h b/source/Lib/CommonLib/CommonDef.h
index 73a4b1fd2f..945ae33f84 100644
--- a/source/Lib/CommonLib/CommonDef.h
+++ b/source/Lib/CommonLib/CommonDef.h
@@ -203,6 +203,11 @@ static const int MINIMUM_TID_ENABLING_TEMPORAL_INPUTS = 3; // JVET-A
#endif
#endif
+#if JVET_AC0196_NNSR
+static const int NNSR_ENABLING_WIDTH = 3840;
+static const int NNSR_ENABLING_HEIGHT = 2160;
+#endif
+
#if JVET_AC0055_NN_POST_FILTERING
static const int MAX_NUM_NN_POST_FILTERS = 8;
static const int NNPF_BLOCK_SIZE = 64;
diff --git a/source/Lib/EncoderLib/EncCfg.h b/source/Lib/EncoderLib/EncCfg.h
index 9136c0a804..85403fd4ef 100644
--- a/source/Lib/EncoderLib/EncCfg.h
+++ b/source/Lib/EncoderLib/EncCfg.h
@@ -176,7 +176,9 @@ protected:
std::string m_rdoCnnlfIntraLumaModelNameNNFilter1; ///< intra luma nnlf set1 model
#endif
#if JVET_AC0196_NNSR
- bool m_nnsrOption;
+ bool m_nnsrOption;
+ int m_nnsrQpOffset;
+ std::string m_sNnsrQpOffsetOverridebyQp;
#endif
int m_iFrameRate;
int m_FrameSkip;
@@ -914,6 +916,10 @@ public:
#if JVET_AC0196_NNSR
bool getUseNnsr() { return m_nnsrOption; };
void setUseNnsr(bool b) { m_nnsrOption = b; };
+ int getNnsrQpOffset() { return m_nnsrQpOffset; };
+ void setNnsrQpOffset(int b) { m_nnsrQpOffset = b; };
+ std::string getNnsrQpOffsetOverridebyQp() { return m_sNnsrQpOffsetOverridebyQp; };
+ void setNnsrQpOffsetOverridebyQp(std::string b) { m_sNnsrQpOffsetOverridebyQp = b; };
#endif
void setProfile(Profile::Name profile) { m_profile = profile; }
diff --git a/source/Lib/EncoderLib/EncLib.cpp b/source/Lib/EncoderLib/EncLib.cpp
index edd4984f3d..94c2a9af12 100644
--- a/source/Lib/EncoderLib/EncLib.cpp
+++ b/source/Lib/EncoderLib/EncLib.cpp
@@ -809,7 +809,8 @@ bool EncLib::encodePrep( bool flush, PelStorage* pcPicYuvOrg, PelStorage* cPicYu
ppsID = m_iGOPRprPpsId;
}
#elif JVET_AC0196_NNSR
- ppsID = ENC_PPS_ID_RPR;
+ if (m_resChangeInClvsEnabled)
+ ppsID = ENC_PPS_ID_RPR;
#else
if( m_resChangeInClvsEnabled && m_intraPeriod == -1 )
{
@@ -2532,8 +2533,27 @@ int EncCfg::getQPForPicture(const uint32_t gopIndex, const Slice *pSlice) const
#if ADAPTIVE_RPR
if (pSlice->getPPS()->getPPSId() == ENC_PPS_ID_RPR)
{
- // adjust QP for rate matching.
- qp -= 0;
+ // adjust QP for rate matching.
+ int baseQp = qp;
+ constexpr int maxQp = 64;
+ std::array<int, maxQp> qpOffsets;
+ qpOffsets.fill(m_nnsrQpOffset);
+
+ // rate adaptation for AI
+ if (getIntraPeriod() == 1)
+ {
+ std::istringstream iss(m_sNnsrQpOffsetOverridebyQp);
+ int iqp, qpOffset;
+ while (iss >> iqp >> qpOffset)
+ {
+ if (iqp < 0 || iqp >= maxQp)
+ {
+ std::cerr << "Error in m_sNnsrQpOffsetOverridebyQp format" << std::endl;
+ }
+ qpOffsets[iqp] = qpOffset;
+ }
+ }
+ qp -= qpOffsets[baseQp];
}
#endif
--
GitLab