Issue on poc derivation after JVET-S0081

[forayr]

After JVET-S0081, prevTid0Pic is defined only if a picture with ph_non_ref_pic_flag equal to 0 has been send. However there is no constraint that the CVSS shall have ph_non_ref_pic_flag equal to 0.
Therefore, prevTid0Pic might not exist when decoding the pictures after the Irap at the beginning of a stream. In this case, predPicOrderCntLsb/Msb are not defined but are used on the following pictures of the sequence to derive PicOrderCntVal.

Should we change the spec to allow prevTid0Pic to be a CVSS with ph_non_ref_pic_flag equal to 1 ?
Or should a CVSS have ph_non_ref_pic_flag equal to 0 ?

[yk]

In the POC derivation process, under certain condition (when vps_independent_layer_flag[ currLayerIdx ] is equal to 1 or there is no picture picA in the current AU with nuh_layer_id equal to layerIdA such that GeneralLayerIdx[ layerIdA ] is in the list ReferenceLayerIdx[ currLayerIdx ], and when ph_poc_msb_cycle_val is not present and the current picture is not a CLVSS picture), prevTid0Pic is set to be the previous picture in decoding order that has nuh_layer_id equal to the nuh_layer_id of the current picture, has TemporalId and ph_non_ref_pic_flag both equal to 0, and is not a RASL or RADL picture.

When the "certain condition" is true, the prevTid0Pic for a current picture could be the CVSS picture. That is (already) allowed.

When the "certain condition" is true, and when prevTid0Pic after the CVSS picture in decoding order does not exist, then I think indeed the CVSS picture needs to be the prevTid0Pic, and it needs to have ph_non_ref_pic_flag equal to 0 (it of course also satisfies other conditions of prevTid0Pic , too). To me that seems implied. I am not sure that we should add such a constraint (at this late stage).

Do others have an opinion herein? Note that the text for VVC v1 needs to be finalized tomorrow Tuesday 7/28.

[hendry197]

I don't think we have to constraint that a CVSS shall have ph_non_ref_pic_flag equal to 0.
Encoder should ensure that POC derivation works. This is by either send POC MSB information or make sure there is prevTid0Pic. A CVSS can have ph_non_ref_pic_flag and for later pics, encoder can rely on sending POC MSB info, not relying on the CVSS for POC derivation. While it is not the best scenario, this is possible.

For me collective constraints that we have so far about POC has already implied the point I said above (i.e., For derivation of PicOrderCntMsb, it shall work; either we have POC MSB info signalled or we have prevTid0Pic present within the MaxPicOrderCntLsb / 2 range.

[yk]

OK, I am closing this ticket now without a change made to the spec.