Custom Query (1442 matches)
Results (22 - 24 of 1442)
Ticket | Resolution | Summary | Owner | Reporter |
---|---|---|---|---|
#26 | fixed | Bug fixes for IMVP | fbossen | simon.oudin@… |
Description |
In TMuC0.5 In TComDataCU.cpp, function : insertCollocated, line 2703 The original purpose of the above <break;> is to avoid multiple insertion. However, from the above code, the cMv is always inserted. The fix uses <return;> instead of<break;>. function : clearMVPCand, lines 3311, 3332 and function : clearMVPCand_one_fourth, line 3171 The test to reduce the list of AMVP is false, when HHI_IMVP is turned on. The test should be done only for the vertical component when HHI_IMVP is turned on. function : clearMVPCand, lines 3290, 3292, 3369 The test to reduce the list of AMVP should be done only for the vertical component when HHI_IMVP is turned on (following the last fix). In consequence, before processing this reduction, the vertical components of the AMVP list which are redundant should be erased, when HHI_IMVP is turned on. This fix works like the function xUniqueMVPCand_one_fourth, line 3072. In TDecEntropy.cpp, In function decodeMVPIdx, lines 1010, 1013, 1082, 1085, 1154, 1157, 1124, 1227, 1295, 1298, 1365, 1369, 1502, 1505 Remplacement of iPartIdx by uiPartAddr for the functions : clearMVPCand_one_fourth and clearMVPCand This parameter is used in these functions for the test: !pcCU->isSkip(uiPartAddr) when HHI_IMVP is turned on. This bug allowed IMVP for SKIP/DIRECT, that isn't correct. In function decodeMVPIdx, lines 1036, 1046, 1108, 1118, 1179, 1189, 1259, 1267, 1320, 1330, 1391, 1401, 1528, 1538 Remplacement of iPartIdx by uiPartAddr for the same test: !pcSubCU->isSkip(uiPartAddr) In TEncSearch.cpp, In function predInterSearch, lines 3809, 3824, 3839, 3854, 3890, 3905, 3938, 3953 Remplacement of iPartIdx by uiPartAddr for the functions : clearMVPCand_one_fourth and clearMVPCand This parameter is used in these functions for the test: !pcCU->isSkip(uiPartAddr) when HHI_IMVP is turned on. This bug allowed IMVP for SKIP/DIRECT, that isn't correct. All these fixes are accessible with the flag HHI_IMVP_HHI in the branche TMuC_hhi-0.5. |
|||
#27 | fixed | ROT index coding bug when ROT = 1 and MDDT = 0 | fbossen | wjhan |
Description |
ROT index is now coded even when ROT=0 and MDDT=1. For example: (when MDDT=1)
Should be:
|
|||
#28 | fixed | Incompatibility, HHI_TRANSFORM_CODING=0, ROT=0 | fbossen | wjhan |
Description |
It seems large loss when HHI_TRANSFORM_CODING=0, ROT=0. Maybe due to the wrong quantization scaling matrix for this case. |