///////////////////////////////////////// tag 1.14.0 * libvvenc: - public interface change: - cts and dts in vvencYUVBuffer and vvencAccessUnit are now signed instead of unsigned integers - deprecated LMCS (will be removed in the future) - enable VUI if aspectRatioIdc, colourPrimaries, or matrixCoefficients are set - many more ARM SIMD optimizations for both NEON and SVE - cmake build changes: - add build option VVENC_FPP_CONTRACT_OFF to ensure encoded bitstreams are bit-identical on different platforms - don't require nlohmann_json in INTERFACE_LINK_LIBRARIES - update nlohmann_json to v3.12.0 - ARM: enable SVE intrinsics by default if compiler supports it - some minor memory optimizations - bug fixes and cleanups - fix some build issues * vvenFFapp: - new config parameter: - Range: set colour range: unknown, limited | mpeg | tv, full | jpeg | pc * vvencapp: - new parameters: - --range: set colour range: unknown, limited | mpeg | tv, full | jpeg | pc * vvenc_unit_test: - add unit tests for many of the functions implemented in SIMD - add --seed parameter - add --testcase/-t to run a subset of unit tests - add --fast parameter for faster CI runs - reduce memory sizes for CI unit testing ///////////////////////////////////////// tag 1.13.1 * libvvenc: - fix errors during library loading on old x86 CPUs associated with invalid instructions - necessitate SSE 4.1 support for x86(_64) architectures - fix rate control bug when using slice type adaptation improving accuracy - fix unaligned load/stores on ARM - allow usage of external nlohmann_json (set VVENC_ENABLE_THIRDPARTY_JSON=SYSTEM in CMake) - added optional unstable api extension allowing passing user data with input frames (off per default, preserving API and ABI compatibility) ///////////////////////////////////////// tag 1.13.0 * libvvenc: - many ARM SIMD optimizations for both NEON and SVE - added parameters - m_GOPQPA: enableed GOP adaptive QP cascade (default: on if QPA off) - m_numParallelGOPs: allows to encode multiple GOPs at once, improving multi-threading for large number of threads (default: off) - m_fga (experimental): enables film grain analysis and embeds FGS SEI in the encoded bitstreams (default: off) - changed parameters: - m_DecodingRefreshType: added a DASH optimized mode VVENC_DRT_IDR_NO_RADL - starts each intra period with an IDR frame without leading picture, allowing simple splitting into segments at each IDR. - preset adjustments: - 5% speedup for medium - 2% speedup for slow and slower - memory reduction for medium/fast/faster - bugfixes - cleanups * vvencFFapp: - added parameters: - GOPQPA: enabled GOP adaptive QP cascade (default: on if QPA off) - NumParallelGOPs: allows to encode multiple GOPs at once (default: off) - FGA: enables the experimental film grain analysis (default: off) - changed parameters: - DecodingRefreshType: now also accepts values "6" and "idr_no_radl" enabling the new DASH optimized IDR only structure. * vvencapp: - changed parameters: - refreshtype: now also accepts values "6" and "idr_no_radl" enabling the new DASH optimized IDR only structure. ///////////////////////////////////////// tag 1.12.1 * libvvenc: - changed parameters (vvenc_config::) for proper VUI signalization: - m_aspectRatioIdc=1 per default (sar 1x1) - m_chromaLocInfoPresent=-1 and m_chromaSampleLocType=-1 per default (auto behavior depending on HDR/SDR mode) m_chromaSampleLocType is auto set, depending on m_horCollocatedChromaFlag and m_verCollocatedChromaFlag m_chromaLocInfoPresent=1 and m_chromaSampleLocType=2 for VVENC_HDR_PQ, VVENC_HDR_PQ_BT2020, VVENC_HDR_HLG, VVENC_HDR_HLG_BT2020, VVENC_SDR_BT2020 else m_chromaLocInfoPresent=0 and m_chromaSampleLocType=0 - m_aspectRatioInfoPresent flag is auto on, if m_aspectRatioIdc>0 or non-default Sar is given - m_overscanInfoPresent flag is auto on, when m_overscanAppropriateFlag is enabled - m_chromaSampleLocTypeTopField and m_chromaSampleLocTypeBottomField unused now (will be removed in next major release) - VUI defaults changed: - progressiveSourceFlag=1 - nonPackedFlag=1 - nonProjectedFlag=1 * vvencFFapp: - added parameters: - Sar: specify Sample aspect ratio - ratio of width to height (WidthxHeight) - removed unused parameter: ChromaSampleLocTypeTopField, ChromaSampleLocTypeBottomField - changed parameter section: - cleanup section "VUI and SEI options" into separate sections: "HDR/SDR and Color Options", "SEI and auxiliary options", "VUI options" - "ChromaLocInfoPresent" is now auto behavior "Signals whether chroma_sample_loc_type is present (-1: auto, 0: off, 1: on)") ///////////////////////////////////////// tag 1.12.0 * libvvenc: - added parameters: - vvenc_config::m_mtProfile: defines how aggressively to boost multi-threading (0...3, where 0 is current multi-threading, and 3 contains all WPP, IFP and tile usage, use -1 for auto behavior depending on resolution and the number of threads used) - changed parameters (vvenc_config::): - m_numTileCols and m_numTileRows from uint32_t to int32_t (-1 for auto behavior, see m_mtProfile) - m_ifp and m_entropyCodingSynchEnabled from bool to int8_t (-1 for auto behavior, see m_mtProfile) - enabled capped constant quality factor (CQF) mode - constant quality mode (vvenc_config::m_QP set and ::m_usePerceptQPA enabled) with rate capping enabled with an absolute limit (::m_RCMaxBitrate) - minor (~5%) speedups to faster preset - improvements to low-delay encoding configuration - other minor optimizations, cleanups and bugfixes * vvencFFapp: - added parameters: - MTProfile: set the multi-threading profile (0: current default - 3: use all MT optimizers, use -1/auto to automatically set depending on resolution and number of threads used) - changed parameters: - WaveFrontSynchro, IFP, Tiles: default value set to -1, with the behavior defined by MTProfile parameter, resolution and number of used threads * vvencapp: - added parameters: - mtprofile: set the multi-threading profile (0: current default - 3: use all MT optimizers, use -1/auto to automatically set depending on resolution and number of threads used) - internal-bitdepth: sets internal encoding bitdepth, i.e. the output bitstream bitdepth - refreshtype: specifies the refresh-type used at random access points - decodedpicturehash: enable/disable decoded picture hash (DPH) SEI generation - changed parameters: - ifp, tiles: default value set to -1, with the behavior defined by mtprofile parameter, resolution and number of used threads - format: now accepts yuv400 and yuv400_10 as parameter for monochrome input ///////////////////////////////////////// tag 1.11.1 * libvvenc: - fix performance regression for ARM - using SSE41 code to generate NEON with SIMDe rather than AVX2 ///////////////////////////////////////// tag 1.11.0 * libvvenc: - added library parameters: - vvenc_config::m_forceScc to force a specific SCC detection decision for all frames (0: do not use, 1/2/3: no SCC/weak SCC/strong SCC) - vvenc_config::m_ifp to enable inter-frame parallelism - vvenc_config::m_minIntraDist to define minimal distance between intra frames (in presentation order) - changed parameters: - vvenc_config::m_RCMaxBitrate can now be specified as factor of target bitrate, use the macro VVENC_SET_MAXRATE_FACTOR to set the value (absolute of negative values will be interpreted as a 4-bit fixed-point factor) - vvenc_config::m_numThreads will now per default be set to 12 for 5K and higher resolutions - renamed library parameters: - vvenc_config::m_fppLinesSynchro to vvenc_config::m_ifpLines - added an affine speedup based on Pejman et al., ICIP 2023. - added vectorized DQ implementation - finalized inter-frame parallelization (previously FPP) improving performance during execution with high number of threads (harmonized with rate control) - improved SCC detection by reducing false positives - overall speedups: - 6% for fast/slow/slower - 12% for medium, 25%+ for multi-threaded execution with long intra periods * vvencFFapp: - added parameters: - IFP: enables the usage of inter-frame parallelism - IFPLines: defines the synchronization offset in CTU lines - ForceSCC: forces a specific SCC detection decision - MinIntraDistance: defines a minimal distance between intra pictures in presentation order - changed parameters: - FppLinesSynchro: now deprecated, please use IFP and IFPLines - NumThreads: if set to default (-1), will now cause the encoder to use 12 threads for 5K and higher resolutions - MaxRate: if specified with suffix "x", e.g. "2x", the value will be interpreted as multiple of target bitrate * vvencapp: - added parameters: - ifp: when set, will enable inter-frame parallelism with a default synchronization offset of two CTU lines - changed parameters: - threads, t: if set to default (-1), will now cause the encoder to use 12 threads for 5K and higher resolutions - maxrate, m: if specified with suffix "x", e.g. "2x", the value will be interpreted as multiple of target bitrate ///////////////////////////////////////// tag 1.10.0 * libvvenc: - added library parameters: - vvenc_config::m_poc0idr to force POC0 to be an IDR (otherwise per default it will now be a RASL picture) - changed default GOP structure of the first GOP to align with other GOPs, ensuring all intra-periods and DASH segments have equal length - enable spatial subsampling for 1st pass in 2-pass RC per default for faster preset - adaptively extend the number of neighboring frames in MCTF pre-filtering - allow ALF derivation from partial data for FPPLinesSynchro - improved rate matching accuracy at high target rates on easy-to-encode content, especially for HDR - disable compilation for SSE42 and AVX as no specific code is used, only keep explicit support for SSE41 and AVX2 - around 0.5% BDR gain for all presets - minor changes and cleanups to rate control, DMVR, MCTF, QPA and others * vvencFFapp: - added parameter: POC0IDR to control if POC0 is an IDR (=1, default if PicReordering is 1) or RASL (=0, default if PicReordering is 0) ///////////////////////////////////////// tag 1.9.1 * libvvenc: - added library parameters: - vvenc_config::m_maxPicWidth, vvenc_config::m_maxPicHeight to signal maximal possible picture size when encoding for resolution change (e.g. using CRA_CRE decoder refresh) - added medium_lowDecEnergy as a built-in preset (in C-API: VVENC_MEDIUM_LOWDECNRG) - refactoring and improvements of affine search and interpolation filtering - fixed 8-bit encoding from 10-bit input and proper handling of 8-bit video in BIM - added proper setting of conformance window - refactoring and improvements to rate control (RC) - fixed potential integer overflow - finalized spatial subsampling implementation for 2-pass RC * vvencFFapp: - added parameter: MaxPicSize/MaxPicWidth/MaxPicHeight to control the signaled maximal picture size during resolution change ///////////////////////////////////////// tag 1.9.0 * libvvenc: - removed library parameters: vvenc_config::m_wcgChromaQpControl, vvenc_config::m_ccalfQpThreshold, vvenc_config::m_switchPOC, vvenc_config::m_switchDQP, vvenc_config::m_fastForwardToPOC, vvenc_config::m_stopAfterFFtoPOC, vvenc_config::m_bs2ModPOCAndType, vvenc_config::m_forceDecodeBitstream1, vvenc_config::m_decodeBitstreams - removed deprecated struct: vvencWCGChromaQPControl - added library parameters: - vvenc_config::m_fppLinesSynchro (controls CTU line offset for dependent frame parallelism) - vvenc_config::m_RCMaxBitrate (controls the maximum instantaneous rate for rate control encoding) - changed library parameters: - vvenc_config::m_FirstPassMode (added spatial and temporal subsampling to first pass) - vvenc_config::m_decodedPictureHashSEIType (when >= 10, the hash is calculated and logged, but no SEI message is generated) - vvenc_config::m_IntraPeriodSec (will now round the intra period to a multiple of 8 frames, instead of GOP size) - added rate capping during rate control (must be at least 1.5x of target rate) - extensive improvements and cleanups to rate control, especially single pass - improved efficiency by 1-2% - speedup of first pass, especially for preset faster (around 15% overall runtime) - unified MCTF activation for low- and high-rate encoding - various speedups, fixes, and cleanups * vvencapp: - added parameter: stats (when enabled, prints a statistical summary of encoding every second of input) maxrate (enables to specify maximum rate for RC, must be at least 1.5x of target rate) * vvencFFapp: - added parameter: MaxBitrate (enables to specify maximum rate for RC, must be at least 1.5x of target rate) stats (when enabled, prints a statistical summary of encoding every second of input) - removed parameters: WCGPPSEnable, WCGPPSCbQpScale, WCGPPSCrQpScale, WCGPPSChromaQpScale, WCGPPSChromaQpOffset, DebugBitstream, DecodeBitstream1, DecodeBitstream2, DebugPOC, SwitchPOC, SwitchDQP, FastForwardToPOC, StopAfterFFtoPOC, DecodeBitstream2ModPOCAndType, ForceDecodeBitstream1, CCALFQpTh, decode ///////////////////////////////////////// tag 1.8.0 * libvvenc: - removed library parameters: - vvenc_config::m_temporalSubsampleRatio, vvenc_config::m_MaxCodingDepth, vvenc_config::m_RCForceIntraQP, vvenc_config::m_log2DiffMaxMinCodingBlockSize - added library parameters: - vvenc_config::m_FirstPassMode - changed library parameters: - vvenc_config::m_useSelectiveRDOQ is now an int_8, with 2 enabling selective RDOQ only for SCC - deprecated config: vvencWCGChromaQPControl - QPA: added saliency detection - BIM: propagation of the parameters to higher temporal layers - retuned the presets: - 15% speedup for faster, 10% speedup for fast, medium and slow, 5% for slower - speedup of the first pass for faster preset - various speedups and bugfixes * vvencapp: * vvencFFapp: - removed from default installation! Use VVENC_INSTALL_FULLFEATURE_APP CMake variable to override. - removed parameters: TemporalSubsampleRatio, RCForceIntraQP, MaxMTTHierarchyDepth, MaxMTTHierarchyDepthI, MaxMTTHierarchyDepthISliceL, MaxMTTHierarchyDepthISliceC - deprecated parameters: WCGPPSEnable, WCGPPSCbQpScale, WCGPPSCrQpScale, WCGPPSChromaQpScale, WCGPPSChromaQpOffset - added parameters: FirstPassMode ///////////////////////////////////////// tag 1.7.0 * libvvenc: - removed (unused) library parameters: - vvencGOPEntry: m_CbTcOffsetDiv2, m_CbBetaOffsetDiv2, m_CrTcOffsetDiv2, m_CrBetaOffsetDiv2 m_refPic, m_isEncoded, m_ltrp_in_slice_header_flag - vvencReshapeCW: rspBaseQP - added library parameters: - extended vvencHDRMode to also include SDR color spaces - vvenc_config: m_numRefPics, m_numRefPicsSCC, m_blockImportanceMapping, m_saoScc - vvenc_config: changed m_sliceTypeAdapt from bool to char - changed library parameters: - m_TicksPerSecond new default 27000000 (old: 90000) to support all NTSC frame rates - added support for ARM through SIMDe - allowing adaptive intra period (content adaptive placement of random access points) - ported block importance mapping from VTM - around 1% gain for faster - around 2% BD-rate gain for other presets - added noise based QP clipping to RC - improved SCC detector to ignore black borders - added possibility of logo overlay - major refactoring to separate a pre-processing step - many bugfixes - speedups: 20% for faster, 5% for other presets * vvencapp: - added parameter "sdr" to set the SDR color space (mutually exclusive with "hdr" parameter) - added parameter "logofile" specifying a JSON logo descriptor (use without value for help) * vvencFFapp: - added parameters "Sdr" and "logofile" analogue to simple app - added parameters "NumRefPics" and "NumRefPicsSCC" allowing per-TL setting of number of reference pictures - added parameter "BIM" - changed parameters "SAO" and "STA" from flag to integer - SAO=2 will only apply SAO to screen content - STA=2 will start a new intra period when a content-dependent intra frame is inserted ///////////////////////////////////////// tag 1.6.1 * libvvenc: - bugfix for rate control at low rates - bugfix to multi-threaded encoding with tiles - made apputils an internal header only lib, fixing linking problems - speedup non-linear ALF (-4% runtime for slower) - added auto-logic for correct tickspersec setting - optimized non-RD-optimized quantization ///////////////////////////////////////// tag 1.6.0 * libvvenc: - removed library parameters: - vvenc_config::m_bRestrictMESampling - added library parameters: - vvencMCTF::MCTFUnitSize, vvenc_config::m_alfUnitSize, vvenc_config::m_meReduceTap - changed semantics of the parameters: - vvenc_config::m_maxMTTDepth: if >=10, each decimal digit corresponds to the value set for frames in a specific temporal layer - vvenc_config::m_maxMTTDepthIChroma: if negative, set equal to vvenc_config::m_maxMTTDepthI - vvenc_config::m_motionEstimationSearchMethod: deprecated value 2 - vvenc_config::m_motionEstimationSearchMethodSCC: deprecated value 1 - speedups - 15%+ for faster, fast, medium and slower - 6% for slow - improvements to rate control stability, especially for single pass and high rates - various improvements, cleanups and fixes * vvencapp: - remove deprecated gopsize parameter * vvencFFapp: - removed command line parameters: - RestrictMESampling - added command line parameters: - ReduceFilterME, ALFUnitSize, MCTFUnitSize ///////////////////////////////////////// tag 1.5.0 * libvvenc: - removed library paramters: - vvenc_config::m_RPLList0, vvenc_config::m_RPLList1, vvenc_config::m_maxDecPicBuffering, vvenc_config::m_maxNumReorderPics, vvenc_config::m_maxNumReorderPics, vvenc_config::m_numRPLList0, vvenc_config::m_numRPLList1, vvenc_config::m_bUseConstrainedIntraPred, vvenc_config::m_deblockingFilterMetric, vvenc_config::m_videoSignalTypePresent - added library parameters: - vvenc_config::m_deblockLastTLayers, vvenc_config::m_picReordering, vvenc_config::m_fastHad - allowing arbitrary intra periods (not required to be multiples of the GOP size) - improvements to rate control stability, mostly regarding to noisy input and scene changes - many speedups and improvements - 10% speedup, 1.3% BD-rate gain for faster - 12% speedup, 1.3% BD-rate gain for fast - 11% speedup for medium - 8% speedup for slow - 6% speedup for slower - added a low-decoding-energy configuration roughly matching the medium preset - fixed portability - multi-threading improvements - memory reduction * vvencapp: * vvencFFapp: - removed command line parameters: - DeblockingFilterMetric, ConstrainedIntraPred - added command line parameters - PicReordering: enable/disable picture reordering during encoding (default 1, picture reordering enabled) - FastHAD: enable/disable the usage of a fast subsampled HAD calculation for large blocks - DeblockLastLayers: indicates to only use deblocking filter for the N highest temporal layers (default 0, deblock all temporal layers) ///////////////////////////////////////// tag 1.4.0 * all: - changed license to the standard Clear BSD license (without adaptations) - explicitly stating the authors (AUTHORS.md) as copyright holders * libvvenc: - removed library parameters: - vvencMCTF::MCTFNumLeadFrames, vvencMCTF::MCTFNumTrailFrames - vvenc_config::m_RCLookAhead (now always behaving as if equal to 1) - added library parameters: - vvenc_config::m_leadFrames, vvenc_config::m_trailFrames, vvenc_config::m_LookAhead, vvenc_config::m_sliceTypeAdapt - add function to extract SPS, PPS and other headers during encoding int vvenc_get_headers(vvencEncoder *, vvencAccessUnit *); - removed legacy 1-pass rate control (now always using the look-ahead based 1pRC) - added a slice adaptation mode inserting intra-frames instead of TL-0 B-frames when scene-cut is detected within the preceding GOP - fix encoder initialization with other bitstream when using rate control - fixed CU tracing - ported improvements to motion compensated filtering from JVET - fixed LMCS processing for 8-bit internal bitdepth coding - various fixes and improvements * vvencapp: - changed semantics of the bitrate parameters to understand suffixes (Mbps, M, kbps, k, bps), associated with multipliers (1e6, 1e6, 1e3, 1e3, 1), respectively - added y4m parameter to force input interpretation as Y4M (also inferred when file file extension is .y4m) * vvencFFapp: - added command line parameters: - STA: enable/disable TL-0 B-to-I slice type adaptation after scene cuts (default: on) - renamed parameters: - MCTFNumLeadFrames -> LeadFrames (now also relevant in chunk based encoding for STA) - MCTFNumTrailFrames -> TrailFrames - changed semantics of the TargetBitrate parameter to understand suffixes (Mbps, M, kbps, k, bps), associated with multipliers (1e6, 1e6, 1e3, 1e3, 1), respectively - added y4m parameter to force input interpretation as Y4M (also inferred when file file extension is .y4m) ///////////////////////////////////////// tag 1.3.1 * libvvenc: - fixed QPA encoding with tiles - allowing encoding of videos compatible with subpic-merging functionality of VVC - API extended, but ABI compatible - added paramter: m_treatAsSubPic, m_explicitAPSid - temporarily forbidding rate control combined with DebugBitstream * vvencapp: - fixed output of `vvencapp --version` - aligned output with full-featured app * vvencFFapp: - aligned output with vvencapp - added command line parameters - TreatAsSubPic - ExplicitAPSid ///////////////////////////////////////// tag 1.3.0 * libvvenc: - introducing params setting using a string api (vvencCfg.h): int vvenc_set_param (vvenc_config *cfg, const char *name, const char *value); int vvenc_set_param_list(vvenc_config *c, int argc, char* argv[]); - logger interface changed to support multiple encoder instances. use: vvenc_set_msg_callback(vvenc_config *, void* , vvencLoggingCallback) - deprecated: vvenc_set_logging_callback(void * ctx, vvencLoggingCallback) (please use: vvenc_set_msg_callback) - added CRA with constrained RASL encoding refresh type (cra_cre) - presets redefined for codec improvements - add look ahead analysis for single pass rate control - HRD timing information are always written into bitstream per default (signal fps) - perceptual QPA improvements for lower bitrates - added tile support including tile parallelism - mingw support added - various cleanups, bugfixes and reducing memory usage * vvencapp: - added command line parameters: - tiles: set number of tile columns and tile rows - framescale: defines the denominator of the framerate to enable fractional rate sepcification (before framerate 59 automatically used 60000/1001) - fps: framerate/framescale in a single parameter as integer or fraction (e.g. 60, 60/1 or 60000/1001) - additional: set parameters using the string api by using a space separated list of key=value parameters (e.g. "bitrate=1000000 passes=2") - rt/refreshtype now supports cra_cre (previously set using --RPR=2 in full-featured app) * vvencFFapp: - added command line parameters: - Tiles: set number of tile columns and tile rows - FrameScale: defines the denominator of the framerate to enable fractional rate sepcification (before framerate 59 automatically used 60000/1001) - fps: framerate/framescale in a single parameter as integer or fraction (e.g. 60, 60/1, 60000/1001) - additional: set parameter using the string api by using a space separated list of key=value parameters (e.g. "bitrate=1000000 passes=2") - LookAhead: enable pre-analysis in single pass rate control - DecodingRefreshType/dr now supports value '5' (cra_cre), previously set using --RPR=2) - removed command line parameters: - InputQueueSize - removed unused/legacy command line parameters (legacy VTM parameter): - EnablePictureHeaderInSliceHeader - CropOffsetLeft, CropOffsetTop, CropOffsetRight, CropOffsetBottom - CalculateHdrMetrics - SignalledIdFlag - SignalledSliceIdLengthMinus1 - RectSlicesBoundaryArray - SignalledSliceId - isSDR ///////////////////////////////////////// tag 1.2.0 * libvvenc: - added parameters: m_qtbttSpeedUpMode, m_numIntraModesFullRD, m_reduceIntraChromaModesFullRD, m_RCPass - renamed parameters: m_bUseEarlyCU to m_useEarlyCU, m_dqThresholdVal to m_quantThresholdVal - added IDR2 refresh type, which aligns the layout of the first intra period with others (using IDR_W_RADL rather than IDR_N_LP picture) - added nlohmann_json as a third party lib (can be disabled with CMake variable VVENC_ENABLE_THIRDPARTY_JSON) - redefined the presets - improvements to single picture rate control - added additional partitioning speedups - improved linear ALF implementation (memory access patterns and SIMD) - improved DMVR implementation - improved error handling - various cleanups and bugfixes * vvencapp: - added command line parameters: - rcstatsfile: defines the location of the to-be-written/to-be-read rate control statistics file - pass: if != -1, defines which encoding pass is to be performed in case of 2-pass RC (requires rcstatsfile) - added support for yuv420_10_packed (packed 10-bit YUV) raw bitstream layout * vvencFFapp: - added command line parameters: - RCStatsFile, Pass (analogue to vvencapp) - NumIntraModesFullRD, ReduceIntraChromaModesFullRD - renamed command line parameters: DQThrVal to QuantThrVal - added support for yuv420_10_packed (packed 10-bit YUV) raw bitstream layout (PYUV=1) ///////////////////////////////////////// tag 1.1.0 * libvvenc: - mostly about faster/fast speedups, minor improvements to other presets - redefined the presets - improved RC rate accuracy - added tool: BCW - added parameters: m_FIMMode, m_bIntegerET, (m_vvencMCTF.)MCTFSpeed, m_alfSpeed, m_IntraEstDecBit, m_BCW, m_motionEstimationSearchMethodSCC - removed parameters: m_useEarlySkipDetection - SCC coding: using alternative ME strategy, disabling LMCS - memory reduction (storing coeffs as 16bit, single storage for displacement/affine motion models, move objects rather than copy during search, garbage collection) - removed dynamic memory from partitioner - SIMD search for last coeff in DQ=0 RDOQ=2 - combined regular and affine merge search into a single function - optimized motion search - RC cleanup - various cleanups * vvencapp: * vvencFFapp: - added options: IntegerET, FastInferMerge, MCTFSpeed, ALFSpeed, IntraEstDecBit, BCW, FastSearchSCC - removed options: ESD * bugfixes: - fixed the TU-split z-partitioning - fixed the ECU option - fixed picture/sequence-level deblocking parameter management - fixed SwitchDQP option when used with multi-threading - fixed TemporalSubsampling (used in JVET AI test conditions) - fixed ALF covariance accumulation domain to double when required (LMCS+ALF+Hdr=pq) - fixed IDR coding with GOPSize==IntraPeriod - fixed short sequence coding (<3 frames) with MCTF ///////////////////////////////////////// tag 1.0.0 * libvvenc: - add pkgconfig - rewriting the vvenc interface in c - restructuring vvenc.h and vvencCfg.h to pure C syntax - removing namespace vvenc and replacing all struct and enumx to typedefs with prefix 'vvenc' - removing file vvencCfgExpert.h and integrating all expert options into single struct vvenc_config - vvencCfg replacing existing lists/vector by fixed size arrays, std::strings are fixed size char arrays now - added IBC - added parameters m_IBCMode (0 - off, 1 - always on, 2 - on with SCC auto detection) and m_IBCFastMethod (1 - slowest with biggest gain to 6 - fastest with lowest gain) - improvements to rate control based on the XPSNR model - supports for generation of main10stillpicture profile bitstreams * vvencapp: - adding parameter --version to show current version - changed parameter: -i/--input accepts "-" to indicate reading input from stdin * vvencFFapp: - adding parameter --version/-v to show current version - vvenc::YUVWriterIf is replaced by callback function vvencRecYUVBufferCallback - changed parameter: -i/--InputFile accepts "-" to indicate reading input from stdin - added parameters IBC and IBCFastMethod as described in the libvvenc * bugfixes - fixes #51, #47, #46, #37, #33 and many more - fixed AVX2 CCALF for non-multiple of 8 picture width - fixed PQ content encoding with LMCS - fixed C++17 compatibility - fixed level/tier auto-settings - fixed FastForwardToPOC and DebugBitstram options ///////////////////////////////////////// tag 0.3.1.0 * libvvenc: - only supporting picture-level rate control, removed vvencCfgExpert::m_RCRateControlMode, vvencCfgExpert::m_RCKeepHierarchicalBit and vvencCfgExpert::m_RCUseLCUSeparateModel parameters. - added vvencCfgExpert::m_log2MinCodingBlockSize parameter to limit minimal coding block size (default: 2) - reduced and defragmented the used memory. - added new speed-ups and adapted fast and faster presets. * vvencFFapp: - removed parameters --KeepHierarchicalBit, --RCLCUSeparateModel and --RateControl (defaults to 2 if target rate is set, 0 otherwise). - renamed parameter --InitialQP to --RCInitialQP. - added parameter --Log2MinCodingBlockSize (default: 2) to limit the minimal coding block size * bugfixes - fixes #29 - corrected motion vector clipping during DMVR process (might produce enc/dec mismatches). - fixed LMCS/ISP interaction (weighted SAD for blocks of width '1'). ///////////////////////////////////////// tag 0.3.0.0 * new: - support for HDR metadata (--hdr off, pq|hdr10, pq_2020|hdr10_2020, hlg, hlg_2020) - if using HDR, encoder decisions depend on the used mode and the enabled metadata. by default VUI and SEI metadata are enabled - PQ: set vui parameter (transfer characteristics 16(smpte2084)) - HLG: set vui parameter (transfer characteristics 1|14(bt709|bt2020-10)) + SEI AlternativeTransferCharacteristics 18 - support for MasteringDisplayColourVolume and MaxContentLightLevel (SMPTE ST 2086 - HDR10/PQ) - MasteringDisplayColourVolume can be set by vector of 10 values: x,y,x,y,x,y,x,y,max,min where: "G(x,y)B(x,y)R(x,y)WP(x,y)L(max,min) x,y coordinates for GBR and WP in increment of 1/50000 max,min luminance in units of 1/10000 cd/m2 0 <= x,y <= 50000, 0 <= max,min <= 10000, Example for a P3D65 1000-nits monitor, where G(x=0.265, y=0.690), B(x=0.150, y=0.060), R(x=0.680, y=0.320), WP(x=0.3127, y=0.3290), L(max=1000, min=0.0001): 13250,34500,7500,3000,34000,16000,15635,16450,10000000,1 - MaxContentLightLevel can be set by vector of 2 values: cll,fall where cll=max. content light level,fall=max. frame average light level, range: 1 <= cll,fall <= 65535 * changes: - align usage of vvenc interface for vvencapp and vvencFFapp (both apps are now using the lib in the same manner) - vvencapp is now using the command line parser of vvencFFapp, but the command line params haven´t changed to prev. version for legacy support - help dialog re-designed, params are now sorted into categories - perceptually motivated QP adaptation (qpa) changed to on/off (mode 0-5 before) and uses XPSNR depending on HDR mode vvencapp: use --qpa or --qpa 0|1 to enable/disable (enabled by default vvencFFapp: use -qpa|--PerceptQPA or -qpa|--PerceptQPA 0|1 to enable/disable (enabled by default - Level is set to auto per default. In this mode it will be automatically detected by given resolution * bugfixes: - #18: setting of max number CABAC-coded coefficients for the floating point scalar RDOQ * libvvenc: - changed preset configuration to reflect current Pareto-Optimal trade-offs. The new presets are much better harmonized between different resolutions and content types. - auto config initialization. The configuration to init the vvenc library now only needs required parameter (width, height, framerate), everything else is set to default config some parameter are set to -1 per default. Params set to -1 will be set to default, auto config params: profile, level, threads, aud, vui, hrd, PerceptQPATempFiltIPic, RateControl, KeepHierarchicalBit, WppBitEqual, FEN - medium preset is now used as default (this has quite a big impact on default params in the vvencFFapp) - decreased memory usage - improved multi-threading (now can utilize up to 10 threads for HD and up to 24 for UHD, efficiency depends on preset and other parameters) - added the possibility to generate bitstreams for minimal drift in open-GOP adaptive switching. - various optimizations and speed-ups. * vvencapp: - new parameter -aud, --accessunitdelimiter [auto] Emit Access Unit Delimiter NALUs (auto(-1),off(0),on(1); default: auto - only if needed by dependent options) -vui, --vuiparameterspresent [auto] Emit VUI information (auto(-1),off(0),on(1); default: auto - only if needed by dependent options) -hrd, --hrdParameterspresent [auto] Emit VUI HRD information (auto(-1),off(0),on(1); default: auto - only if needed by dependent options) --hdr [off] set HDR mode (+SEI messages) + BT.709 or BT.2020 color space. use: off, pq|hdr10, pq_2020|hdr10_2020, hlg, hlg_2020 - changed parameter --qpa [on] Enable perceptually motivated QP adaptation, XPSNR based (0:off, 1:on) as legacy support the previous modes 1-5 are now mapped to on/1 * vvencFFapp: - new parameter --MaxParallelFrames [-1] Maximum number of frames to be processed in parallel (-1: auto depends on number of threads, 0:off, >=2: enable parallel frames) --hdr [off] set HDR mode (+SEI messages) + BT.709 or BT.2020 color space. use: off, pq|hdr10, pq_2020|hdr10_2020, hlg, hlg_2020 --MasteringDisplayColourVolume [[]] SMPTE ST 2086 mastering display colour volume info SEI (HDR), vec(uint) size 10, x,y,x,y,x,y,x,y,max,min where: "G(x,y)B(x,y)R(x,y)WP(x,y)L(max,min)"range: 0 <= GBR,WP <= 50000, 0 <= L <= uint; GBR xy coordinates in increment of 1/50000, min/max luminance in units of 1/10000 cd/m2 --MaxContentLightLevel [[]] Specify content light level info SEI as "cll,fall" (HDR) max. content light level, max. frame average light level, range: 1 <= cll,fall <= 65535' --PreferredTransferCharacteristics [auto] Specify preferred transfer characteristics SEI and overwrite transfer entry in VUI (0-18): reserved, bt709, unknown, empty, bt470m, bt470bg, smpte170m, smpte240m, linear, log100, log316, iec61966-2-4, bt1361e, iec61966-2-1, bt2020-10, bt2020-12, smpte2084, smpte428, arib-std-b67 --segment [off] when encoding multiple separate segments, specify segment position to enable segment concatenation (first, mid, last) [off] first: first segment mid : all segments between first and last segment last : last segment --FastIntraTools [0] SpeedUPIntraTools:LFNST,ISP,MTS. (0:off, 1:speed1, 2:speed2) - changed parameter --qpa [off] Enable perceptually motivated QP adaptation, XPSNR based (0:off, 1:on) -aud, --AccessUnitDelimiter [auto] Enable Access Unit Delimiter NALUs, (default: auto - enable only if needed by dependent options) -vui, --VuiParametersPresent [auto] Enable generation of vui_parameters(), (default: auto - enable only if needed by dependent options) -hrd, --HrdParametersPresent [auto] Enable generation of hrd_parameters(), (default: auto - enable only if needed by dependent options) --Profile [auto] name to use for encoding. Use [multilayer_]main_10[_444][_still_picture], auto, or none --Level [auto] limit to be used, eg 5.1, or none --WppBitEqual [-1] Ensure bit equality with WPP case (0:off (sequencial mode), 1:copy from wpp line above, 2:line wise reset) --RateControl [-1] enable rate control (0:off 1:CTU-level RC; 2:picture-level RC; 3:GOP-level RC) --KeepHierarchicalBit [-1] Rate control: (0:equal bit allocation, 1:fixed ratio bit allocation, 2:adaptive ratio bit allocation --PerceptQPATempFiltIPic [-1] Temporal high-pass filter in QPA activity calculation for key pictures (0:off, 1:on, 2:on incl. temporal pumping reduction, -1:auto) --SliceChromaQPOffsetPeriodicity [-1] Used in conjunction with Slice Cb/Cr QpOffsetIntraOrPeriodic. Use 0 (default) to disable periodic nature. --ColourPrimaries [unknown] Specify color primaries (0-13): reserved, bt709, unknown, empty, bt470m, bt470bg, smpte170m, smpte240m, film, bt2020, smpte428, smpte431, smpte432 --TransferCharacteristics [unknown] Specify opto-electroni transfer characteristics (0-18): reserved, bt709, unknown, empty, bt470m, bt470bg, smpte170m, smpte240m, linear, log100, log316, iec61966-2-4, bt1361e, iec61966-2-1, bt2020-10, bt2020-12, smpte2084, smpte428, arib-std-b67 --MatrixCoefficients [unknown] Specify color matrix setting to derive luma/chroma from RGB primaries (0-14): gbr, bt709, unknown, empty, fcc, bt470bg, smpte170m, smpte240m, ycgco, bt2020nc, bt2020c, smpte2085, chroma-derived-nc, chroma-derived-c, ictcp --RPR [1] can be now set to "2", in which case the encoder will impose constraints on the RASL frames allowing for minimal drift open-GOP adaptive switching (with or withour resolution change). ///////////////////////////////////////// tag 0.2.1.0 * bugfixes: - ISP fix: erroneous placed sanity check * libvvenc: - decrease memory consumption - harmonize 2-pass rate control and perceptual QPA - improve rate control ///////////////////////////////////////// tag 0.2.0.0 * new license - switched to modified 3-clause BSD * bugfixes: - access unit delimiter (AUD) generation - DPH-SEI syntax * new features: - ISP - TS (with automatic screen-content detection) - BDPCM - 2-pass rate control - 1-pass rate control with GOP32 * libvvenc: - added "slower" preset - redefined the presets according to current pareto-set - removed PredictionUnit type - various memory reductions (Rom.cpp, scaling list memory) - verious optimizations (SIMD for MCTF, forward transformation, single column IF; memory accesses for DMVR) - changed MCTF algorithm to do intermediate rounding between hor/ver filtering * vvencapp: - new parameter --refreshsec,-rs to define the intra refresh rate in seconds depending on the given frame rate. Internally, the refresh rate in seconds is translated into the frames where the refresh is set. --internal-bitdepth to define the internal bit-depth used in bitstream (default: 10). --passes to control number of passes for rate-control --segment to control chunkwise encoding configuration - changed parameter --intraperiod,-ip is 0 per default because --refreshsec is used for that purpose instead. In case of a value greater than 0, intraperiod overwrites refreshsec. Abbreviation -q is now supported in addition to --qp. --gopsize default changed to 32 - fullhelp changed All internal encoding parameters are now shown as strings when --fullhelp is called. * build system: - using GnuInstallDirs - added basic tests - versioning using CMake - added support for address sanitizer ///////////////////////////////////////// tag 0.1.0.1 * libvvenc: - VVEnc::init() correctly initializes SIMD for transformation (TrafoX86) and sample operations (BufferX86). * vvencapp: - Improved speed because of the bugfix to libvvenc. ///////////////////////////////////////// tag 0.1.0.0 * initial version /////////////////////////////////////////