Opened 9 years ago

Closed 9 years ago

#1349 closed defect (fixed)

HM cannot decode the 1st buffering period SEI message.

Reported by: gaow02 Owned by: karlsharman
Priority: major Milestone: HM-16.5
Component: HM Version: HM-16.2
Keywords: Cc: ksuehring, davidf, karlsharman, jct-vc@…

Description

We found that current HM trunk cannot correctly decode the buffering period SEI message coded along with the 1st IDR picture.

The code can be fixed as following:

After Fix:

Void TDecTop::xDecodeSEI( TComInputBitstream* bs, const NalUnitType nalUnitType )
{
  //HUA_FIX: the 1st buffering period SEI cannot be correctly decoded if w/o the following fix.
  TComSPS* sps = m_parameterSetManagerDecoder.getActiveSPS();

  if(sps == NULL)
  {
      sps = m_parameterSetManagerDecoder.getPrefetchedSPS(0);
  }

  if(nalUnitType == NAL_UNIT_SUFFIX_SEI)
  {
    m_seiReader.parseSEImessage( bs, m_pcPic->getSEIs(), nalUnitType, sps, m_pDecodedSEIOutputStream );
  }
  else
  {
    m_seiReader.parseSEImessage( bs, m_SEIs, nalUnitType, sps, m_pDecodedSEIOutputStream );

    SEIMessages activeParamSets = getSeisByType(m_SEIs, SEI::ACTIVE_PARAMETER_SETS);
    if (activeParamSets.size()>0)
    {
      SEIActiveParameterSets *seiAps = (SEIActiveParameterSets*)(*activeParamSets.begin());
      m_parameterSetManagerDecoder.applyPrefetchedPS();
      assert(seiAps->activeSeqParameterSetId.size()>0);
      if (! m_parameterSetManagerDecoder.activateSPSWithSEI(seiAps->activeSeqParameterSetId[0] ))
      {
        printf ("Warning SPS activation with Active parameter set SEI failed");
      }
    }
  }
}

Before Fix:

Void TDecTop::xDecodeSEI( TComInputBitstream* bs, const NalUnitType nalUnitType )
{
  if(nalUnitType == NAL_UNIT_SUFFIX_SEI)
  {
    m_seiReader.parseSEImessage( bs, m_pcPic->getSEIs(), nalUnitType, m_parameterSetManagerDecoder.getActiveSPS(), m_pDecodedSEIOutputStream );
  }
  else
  {
    m_seiReader.parseSEImessage( bs, m_SEIs, nalUnitType, m_parameterSetManagerDecoder.getActiveSPS(), m_pDecodedSEIOutputStream );

    SEIMessages activeParamSets = getSeisByType(m_SEIs, SEI::ACTIVE_PARAMETER_SETS);
    if (activeParamSets.size()>0)
    {
      SEIActiveParameterSets *seiAps = (SEIActiveParameterSets*)(*activeParamSets.begin());
      m_parameterSetManagerDecoder.applyPrefetchedPS();
      assert(seiAps->activeSeqParameterSetId.size()>0);
      if (! m_parameterSetManagerDecoder.activateSPSWithSEI(seiAps->activeSeqParameterSetId[0] ))
      {
        printf ("Warning SPS activation with Active parameter set SEI failed");
      }
    }
  }
}

Note that this may not be the best way to fix the issue. But it does help to avoid the decoder crash and allow it to correctly decode the bitstream.

Change History (5)

comment:1 Changed 9 years ago by DefaultCC Plugin

  • Cc ksuehring davidf karlsharman jct-vc@… added

comment:2 Changed 9 years ago by ksuehring

According to the text the SEIs refer to the currently active SPS. If there is no parameter set active, parsing has to be delayed (see section D.3.3 Note 1).

An active parameter sets SEI message can activate the proper SPS before parsing the buffering period (or picture timing) SEI.

The fix assumes, that am SPS with sps_seq_parameter_set_id equal to 0 exists and will be activated in the following VCL NAL unit. Although this assumption is true for all HM generated streams, there is no requirement for this in the bitstream syntax.

comment:3 Changed 9 years ago by gaow02

Please refer to our embedded comments:

According to the text the SEIs refer to the currently active SPS. If there
is no parameter set active, parsing has to be delayed (see section D.3.3
Note 1).
[Comment]: The issue w/ the current HM decoder is that: the parsing of buffering period and picture timing SEI are not properly delayed after the active parameter sets are claimed later by a slice header. It will just try to decode any SEI right at the point of reading it, and if no active sequence parameter set found at the point, it will report an error and exit. So, this operation is not conformant w/ the HEVC spec.

An active parameter sets SEI message can activate the proper SPS before
parsing the buffering period (or picture timing) SEI.

[Comment]: From spec, this is a valid option, but definitely not a requirement for the proper use w/ buffering period (or picture timing) SEIs.

The fix assumes, that am SPS with sps_seq_parameter_set_id equal to 0
exists and will be activated in the following VCL NAL unit. Although this
assumption is true for all HM generated streams, there is no requirement
for this in the bitstream syntax.

[Comment]: As said and identified, the reported fix may not be a good fit for the spec. But as mentioned earlier, an appropriate fix of delaying the BP and PC SEI decoding after slice header decoding, which is really fully conforming to the spec, is needed to clear this un-appropriate exiting issue, when decoding streams w/ BP and SEI messages.

comment:4 Changed 9 years ago by ksuehring

  • Milestone set to HM-16.5
  • Owner set to karlsharman
  • Status changed from new to assigned

comment:5 Changed 9 years ago by ksuehring

  • Resolution set to fixed
  • Status changed from assigned to closed

A fix has been implemented that delays parsing of all prefix SEI messages, until VPS/SPS/PPS are activated by the first VCL NAL unit of the picture.

The change has been committed in r4394

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

  • David Flynn(Subscriber)
  • jct-vc@…(Subscriber)
  • Karl Sharman(Owner, Subscriber)
  • karl.sharman@…(Always)
  • Karsten Suehring(Subscriber, Participant, Always)
  • Wen Gao(Reporter, Participant)