Mismatch spec/VTM on RefPicLtPocList set

[audrey.turquin]

The calculation of FullPocLt[i][j] is not aligned with the specification.
In the VTM, at each calculation (present at multiple places), there is a condition that is not present in the spec.

For example (source/Lib/CommonLib/Slice.cpp, function 'constructRefPicList') :

int ltrpPoc = m_pRPL0->getRefPicIdentifier(ii) & pocMask;
if(m_localRPL0.getDeltaPocMSBPresentFlag(ii))
{
  ltrpPoc += getPOC() - m_localRPL0.getDeltaPocMSBCycleLT(ii) * (pocMask + 1) - (getPOC() & pocMask);
}

The condition "if(m_localRPL0.getDeltaPocMSBPresentFlag(ii))" should not be here, the calculation should be in any case :

int ltrpPoc = m_pRPL0->getRefPicIdentifier(ii) + getPOC() - m_localRPL0.getDeltaPocMSBCycleLT(ii) * (pocMask + 1) - (getPOC() & pocMask);

Moreover, the POC calculated this way is always the POC with MSB.
Consequently, in function xGetLongTermRefPic, the part with "pocHasMsb" is wrong and should be removed.

[audrey.turquin]

Corresponding MR:
​https://vcgit.hhi.fraunhofer.de/jvet/VVCSoftware_VTM/-/merge_requests/1555

[bheng]

I believe the software is working as intended. In the text JVET-R2001-v3, the condition appears in Equation (204). Specifically, the FullPocLt is only used in the "else" case below, otherwise just POC LSB is used.

if( !delta_poc_msb_cycle_lt[ i ][ k ] ) {

...
RefPicLtPocList[ i ][ j ] = PocLsbLt[ i ][ k ]

} else {

...
RefPicLtPocList[ i ][ j ] = FullPocLt[ i ][ k ]

}

However, in the text, I believe this if-condition above was supposed to be based on the flag delta_poc_msb_cycle_present_flag, not the value delta_poc_msb_cycle_lt.

Using delta_poc_msb_cycle_lt in this if-condition doesn't make sense. So, I would suggest changing this to a spec ticket instead, then it should match the software.

[audrey.turquin]

Yes, you are right, I missed this part when setting the RefPicLtPocList. I thought it was always equal to FullPocLt.

Consequently there is only a mismatch about the condition :

• spec uses delta_poc_msb_cycle_lt
• VTM uses delta_poc_msb_cycle_present_flag.

If you say it is rather a spec ticket, I will change it.

[yk]

Good catch! Thanks! Will be fixed in JVET-R2001-v8 (as suggested above by Brian).