Opened 12 years ago Closed 12 years ago #597 closed defect (fixed)encoder-only fix for rdFactor calculation for HE10 settings
Description
The variable ‘rdFactor’ for computing rd-cost in the RDOQ function is currently given by
Int rdFactor = (Int)((Double)(g_invQuantScales[m_cQP.rem()]*g_invQuantScales[m_cQP.rem()]<<(2*m_cQP.m_iPer))/m_dLambda/16 + 0.5) ;
The calculation is based on 8-bit coding (g_uiBitIncrement = 0). It should be computed as follows, so that it also works for the HE10 setting where g_uiBitIncrement = 2.
Int rdFactor = (Int)((Double)(g_invQuantScales[m_cQP.rem()])*(Double)(g_invQuantScales[m_cQP.rem()])*(Double)(1<<(2*m_cQP.m_iPer))/m_dLambda/16/(Double)(1<<(2*g_uiBitIncrement)) + 0.5);
A patch is provided to fix the rdFactor calculation. This fix only affects HE10 settings when RDOQ is enabled. With this fix, on average (-0.13%, -0.32%, -0.54%) BD-rate gain were observed for AI-HE10, RA-HE10 and LB-HE10 settings. Attachments (1)Change History (4)comment:1 Changed 12 years ago by DefaultCC Plugin
Changed 12 years ago by wangjcomment:2 Changed 12 years ago by pandrivoncomment:3 Changed 12 years ago by fbossen
Fixed in r2481 Note: See
TracTickets for help on using
tickets. | This list contains all users that will be notified about changes made to this ticket. These roles will be notified: Reporter, Owner, Subscriber, Participant
|
Thank you for your fix!
Ticket #596 should also be taken into account for 14-bit as coding gains (non-normative technique) with the overflow fixes proposed in ticket #596 (patch over HM7.1rc1) are with class B SVT contents for AI-HE 14-bit (-5.0%, -7.2%, -8.2%) and for RA-HE 14-bit (-2.4%, -5.1%, -2.3%) in comparison with JCTVC-I0108 tests. Tests on 8, 10, 12, 14-bit now show that results are coherent for 14-bit coding.