Custom query (96 matches)


Show under each result:

Results (88 - 90 of 96)

Ticket Resolution Summary Owner Reporter
#88 fixed Missing part_mode binaryzation when quad-tree limitation enabled tech Tomohiro Ikai

When lim_qt_pred_flag is equal to 1, the quad-tree limitation is used. Then the modified part_mode binaryzation and PartMode derivation is used.

The latest spec (JCT3V-I1001_v3) defines the PartMode derivation part's modification (Table I 3 – Name association to prediction mode and partitioning type). However part_mode binaryzation's modification seems missing in the spec as follows:

I. Binarization process for part_mode The specifications in subclause apply.

#89 fixed ARP refIdx and refcheck tech Tomohiro Ikai

At the strasburg meeting, ARP mismatch between text and software was reported. The issues are mainly related to F0105, F0123 and I0051. This ticket reports the details and a suggested fix (relative to HTM-12.2 and J1001_v3). The suggested fix doesn't affect coding results in CTC. The suggested patch includes the four macros (FIX_TICKET_89_REFIDX, FIX_TICKET_89_REFCHECK, FIX_TICKET_89_REFCLEAN, and FIX_TICKET_89_REFCLEAN2).

Issues: (1)Software issue(FIX_TICKET_89_REFIDX):

  • typos in arpIdx selection

(2)Software and text issue(FIX_TICKET_89_REFCHECK):

  • wrong or missing availablity checks in inter-view ARP

Specifically, in inter-view ARP, the ARP reference list X is replaced to Z as proposed in F0123 and I0051, so the software and text should be modified to check the Z rather than X. That's the reasonable harmonization with F0105.

Additionally, a software clean up is provided (FIX_TICKET_89_CLEAN and FIX_TICKET_89_CLEAN2).

Note: FIX_TICKET_89_CLEAN removes unnecessary conditions (2Nx2N check and depth chech are processed before ARP function in text and software, so we can remove them) and removes non-used variable (pNewMvFiled). FIX_TICKET_89_CLEAN2 merges uni-prediction part and bi-prediction part into one part as text specifies. The CLEAN2 changes can be large but it really helps cleanup.

#97 fixed AMVP view MV scaling tech Tomohiro Ikai

View base motion vector scaling in AMVP has been removed (this decision is related to D0122, see the note). But in SW, the scaling part remains. Thus we need to remove the part to solve the text-SW mismatch. A patch relative to r1167 is attached.

Note: This bug was found when confirming/checking 3D-HEVC text of JCT3V-K1001.

Note: See TracQuery for help on using queries.