JVET-AO0010 JVET AHG report: Encoding algorithm optimization (AHG10) [K. Andersson, P. de Lagrange, A. Duenas (co-chairs), T. Ikai, T. Solovyev, A. Tourapis (vice chairs)]
Activity
The regular JVET e-mail reflector was used for discussions (jvet@lists.rwth-aachen.de). No e-mail was sent related to AHG10.
Related contributions
Following contributions were identified relating to AHG10 and summarized in the following sections.
- JVET-AO0136: AHG3/AHG10: On the GOP-based-RPR decision source
This document reports a fix to the VTM reference software to use MCTF filtered samples for GOP based RPR decisions instead of original samples. The issue was introduced unintentionally as part of memory optimizations of VTM.
The fix has already been merged to VTM and the BDR impact when GOP based RPR is enabled for QP 32 to 47 are as follows:
RA:-0,51%(Y),-0,10%(U), -0,18%(V); 95,4%(EncT) 88,8% (DecT), 91,7%(EncVMP),92,6% (DecVMP)
LB:-0,03%(Y),-0,06%(U), -0,08%(V); 99,4%(EncT)100,2%(DecT),99,7%(EncVMP),100,0% (DecVMP)
For CTC (GOP based RPR not enabled in CTC):
RA:0,00%(Y),0,00%(U),0,00%(V);100,6%(EncT),100,0%(DecT),111,0%(EncVMP),101,2%(DecVMP)
LB:0,00%(Y),0,00%(U), 0,00%(V);100,0%(EncT),100,7%(DecT),106,3%(EncVMP),101,4%(DecVMP)
The contribution recommends checking if a similar unintended change in behaviour exists in any other SW packages derived from VTM, e.g. NNVC or ECM, and possibly align the software’s.
Cross-checks in JVET-AO0219 and JVET-AO0266.
- JVET-AO0220: AHG12: Optimizations for JVET-AN0271 software
This document reports some optimizations to the JVET-AN0271 (Software optimization and complexity reduction of ECM). The modification relates to encoder only improvements from the JVET-AN0267 (Ericsson, Fraunhofer HHI and Nokia responses to Joint CfE on Video Compression with Capability beyond VVC) namely 16-tap filter for MCTF, BIM using lambda modification and disabling LMCS for SDR. Also, some bugfixes related to RPR, IntraNN, IntraTMP, PDPC, TIMD, ALF, etc.
The BDR impact compared to JVET-AN0271 is as follows:
CTC RA: -1.27% (Y), -3.97% (U), -3.73% (V); 97.6% (EncT), 100.4% (DecT),
CfE RA: -1.77% (Y), -5.58% (U), -5.46% (V); 97.5% (EncT), 100.5% (DecT).
The document also reports the results in comparison with VTM-23.13 for CfE (not rate matched):
CfE RA: -21.60%(Y), -36.07%(U), -41.35%(V); 1282.3% (EncT), 764.4% (DecT)
CfE LDB: -16.63%%(Y), -33.51%(U), -29.43%(V); 694.7% (EncT), 482.4% (DecT)
Additionally, also with GOP based RPR enabled on Test (not rate matched):
CfE RA: -23.18%(Y), -37.92%(U), -43.10%(V); 1027.9% (EncT), 515.9% (DecT)
CfE LDB: -17.91%(Y), -37.08%(U), -32.98%(V); 652.4% (EncT), 417.6% (DecT)
Recommendations
The AHG recommended that the related input contributions are reviewed, and to further continue the study of encoding algorithm optimizations in JVET.
It was commented that in JVET-AN0267 and JVET-AN0271 some elements might also be beneficial for an VTM encoder. It was suggested that an appropriate way to proceed would be to make a related proposal for VTM optimization.