Custom query (96 matches)
Results (82 - 84 of 96)
| Ticket | Resolution | Summary | Owner | Reporter |
|---|---|---|---|---|
| #49 | fixed | Mismatch in VSP horSplitFlag derivation between WD and HTM | tech | Tomohiro Ikai |
| Description |
The sign of inequality is different between WD(">") and HTM("<"), it poses different results (please see the following detail). Because the mismatch was occured in the text editing phase and the HTM has been tested for some meeting cycles. I think WD fix might be appropriate. An example of WD fix is attached. // WD
horSplitFlag = ( refDepPels[ xP0 ][ yP0 ] > refDepPels[ xP1 ][ yP1 ] )
= = ( refDepPels[ xP1 ][ yP0 ] > refDepPels[ xP0 ][ yP1] ) ) (H 296)
– The variables nSubBlkW and nSubBlkH are modified as specified in the following:
nSubBlkW = horSplitFlag ? 8 : 4 (H 297)
nSubBlkH = horSplitFlag ? 4 : 8 (H 298)
// HTM
ULvsBR = refDepthTmp[0][x+offset[0]] < refDepthTmp[3][x+offset[3]];
URvsBL = refDepthTmp[0][x+offset[3]] < refDepthTmp[3][x+offset[0]];
if( ULvsBR ^ URvsBL )
{ // 4x8
// Detail
------------------------------
WD HTM
------------------------------
> 1 0
== 0 0
< 0 1
------------------------------
--------------------------------------
WD UR>BL UR==BL UR<BL
----------------------------------------
UL>BR 8x4(1==1) 4x8(1==0)* 4x8(1==0)
UL==BR 4x8(0==1)* 8x4(0==0) 8x4(0==0)*
UL<BR 4x8(0==1) 8x4(0==0)* 8x4(0==0)
----------------------------------------
--------------------------------------
HTM UR>BL UR==BL UR<BL
----------------------------------------
UL>BR 8x4(0^0) 8x4(0^0)* 4x8(0^1)
UL==BR 8x4(0^0)* 8x4(0^0) 4x8(0^1)*
UL<BR 4x8(1^0) 4x8(1^0)* 8x4(1^1)
----------------------------------------
|
|||
| #69 | fixed | Missing of DcVal process and unclear DcOffset usage | tech | Tomohiro Ikai |
| Description |
In I.8.4.4.3 (I.8.4.4.3 Segmental depth intra coding process) in 3D-HEVC Draft Text 4 (H1001_v2), dcPred is derived but not used. It seems dcPred related process for recSamples is missing. The fix is not clear but the attached H1001_v2_opt1 may help. To me, the DcOffset usage is not clear in the following points. So the editorial improvement and some fix is to be desired. (1) In SDC, DcOffset is not added to predSamples in I.8.4.4.2.9 but it is added in I.8.4.4.3. The reason why we should have two DcOffset related subclause is not clear to me. If we uses I.8.4.4.2.9 in both cases, the text can be much clear (See H1001_v2_opt2) (2) In SDC, the variable DcVal is derived and used. But the derivation of DcVal is not clear. Specifically, When SDC is on, dcOffsetAvailFlag is false. Then DcOffset is not used both in DLT on case and in DLT off case. But when DLT is on, possibly we should add DcOffset before Idx2DepthValue[] (dcOffsetAvail should be true in this part?). (3) DLT handling is separately specified in I.8.4.4.2.9 and I.8.4.4.3. That is difficult to understand so it should be specified either in I.8.4.4.3 or in I.8.4.4.2.9. (4) If predModeIntra is not equal to INTRA_DMM_WFULL or INTRA_DMM_CPREDTEX, dcPred[0] is derived. But it is unclear whther DcOffset should be added in this case. |
|||
| #75 | fixed | Discrepancy in VSP bi-pred handling | tech | Tomohiro Ikai |
| Description |
There seems to be a discrepancy between spec and software. The spec reads that bipred prevention is applied regardless of VSP flag:
refIdxL1 = −1 (I 112) predFlagL1 = 0 (I 113) However, the software (in TDecEntropy.cpp and TEncSearch.cpp)disbles it in the case VSP flag is true as follows. #if H_3D_VSP
->getVSPFlag(uiSubPartIdx) == 0)) #else
#endif We should remove it (spec should be right). In CTC and IBP(with MTK_I0072_IVARP_SCALING_FIX) experiment, there are no differences by this fix. No change is quite natural because VSP PU cannot be bi-prediction in the current specification. If looking at predFlagLXVSP setting logic(See I.8.5.3.2.13 Derivation process for a view synthesis prediction merge candidate), when predFlagL0VSP is 1, predFlagL1VSP cannot be 1. When predFlagL0VSP is 0, the PU is not bi-prediction. The same logic is applied in VSP inhericance case. Lastly, even if VSP was able to be bi-prediction, we should prevent 8x4/4x8 bipred in VSP as well to preserve motion data storing implemenation. |
|||
