I noticed that simulations which I ran for the ACP format take much more memory than for the other formats. I was using 360Lib-3.0rc1 with HM-16.15. I also did simulations for CMP and ERP. For the same sequence CMP needed 5GB of RAM while ACP needed 12GB. Detailed calls of encoders below.

CMP: bin/TAppEncoderStatic -c cfg/encoder_randomaccess_main10.cfg --QP=37 --SEIDecodedPictureHash=0 --OutputBitDepth=10 -c cfg-360Lib/encoder_360_CMP.cfg --SphFile=cfg-360Lib/360Lib/sphere_655362.txt --Level=5.2 -c cfg-360Lib/per-sequence/360/360test_SkateboardInLot_DynamicViewports.cfg --CodingFaceWidth=1184 --CodingFaceHeight=1184 --IntraPeriod=32 --FrameRate=30 --SourceHeight=4096 --SourceWidth=8192 --InputFile=orig/SkateboardInLot_8192x4096_30fps_10bit_420pf_erp.yuv --BitstreamFile=str/SkateboardInLot_3552x2368_30fps_10bit_420pf_QP37.hevc --FramesToBeEncoded=300 --InputBitDepth=10 --InputChromaFormat=420 --ReconFile=rec/SkateboardInLot_3552x2368_30fps_10bit_420pf_QP37_rec_enc.yuv

maxvmem 5.086G

ACP: bin/TAppEncoderStatic -c cfg/encoder_randomaccess_main10.cfg --QP=37 --SEIDecodedPictureHash=0 --OutputBitDepth=10 -c cfg-360Lib/encoder_360_ACP.cfg --SphFile=cfg-360Lib/360Lib/sphere_655362.txt --Level=5.2 -c cfg-360Lib/per-sequence/360/360test_SkateboardInLot_DynamicViewports.cfg --CodingFaceWidth=1184 --CodingFaceHeight=1184 --IntraPeriod=32 --FrameRate=30 --SourceHeight=4096 --SourceWidth=8192 --InputFile=orig/SkateboardInLot_8192x4096_30fps_10bit_420pf_erp.yuv --BitstreamFile=str/SkateboardInLot_3552x2368_30fps_10bit_420pf_QP37.hevc --FramesToBeEncoded=300 --InputBitDepth=10 --InputChromaFormat=420 --ReconFile=rec/SkateboardInLot_3552x2368_30fps_10bit_420pf_QP37_rec_enc.yuv

maxvmem 12.319G

TDecCABAC.cpp and TEncCABAC.cpp

Old:

#if JVET_F0031_RMV_REDUNDANT_TRSKIP if (isLuma(component) && pcCU->getEmtCuFlag(uiAbsPartIdx))

{

return;

}

#endif

New

#if JVET_F0031_RMV_REDUNDANT_TRSKIP && COM16_C806_EMT

if (isLuma(component) && pcCU->getEmtCuFlag(uiAbsPartIdx)) {

return;

}

#endif

The current 360Lib code (360Lib-3.0rc1) with HM-16.15 while converting from ERP to RVP, we observe PSNR is not calculated properly may be due to not setting the bit depths as SPSNR_NN is disabled for RVP. We think the following might be a bug.

Old code: At line 2119 in source\App\utils\TApp360Convert\TApp360ConvertCfg.cpp

#if SVIDEO_SPSNR_NN

cSPSNRCalc.setSPSNREnabledFlag(m_psnrEnabled[METRIC_SPSNR_NN]); if(m_psnrEnabled[METRIC_SPSNR_NN]) {

cPSNRCalc.setOutputBitDepth(m_outputBitDepth); cPSNRCalc.setReferenceBitDepth(m_referenceBitDepth); cSPSNRCalc.setOutputBitDepth(m_outputBitDepth); cSPSNRCalc.setReferenceBitDepth(m_referenceBitDepth);

}

#endif

Suggested new code: #if SVIDEO_SPSNR_NN

if (m_psnrEnabled[METRIC_PSNR]) {

cPSNRCalc.setOutputBitDepth(m_outputBitDepth); cPSNRCalc.setReferenceBitDepth(m_referenceBitDepth);

}

cSPSNRCalc.setSPSNREnabledFlag(m_psnrEnabled[METRIC_SPSNR_NN]); if(m_psnrEnabled[METRIC_SPSNR_NN]) {

cSPSNRCalc.setOutputBitDepth(m_outputBitDepth); cSPSNRCalc.setReferenceBitDepth(m_referenceBitDepth);

}

#endif