Hello,

I have two points concerning the candidate selection for intra prediction mode.

1. In 8.4.2 (derivation process for luma prediction mode), when candIntraPredModeA == candIntraPredModeB and candIntraPredModeA >= 2,

(8-20) reads candModeList[2] = 2 + ((candIntraPredModeA - 2 + 1) % 32)

so when candIntraPredModeA == 33, this formula gives

candModeList[2] = 2 + ((33 - 2 + 1) % 32) = 2 + (32 % 32) = 2

I think that in this case, we should use the mode 34 for candModeList[2] instead of 2 because it is the neighbouring mode of 33.

If this suggestion is adopted, the new formula for (8-20) will be :

candModeList[2] = 3 + ((candIntraPredModeA - 2) % 32)

2. Also in the same case candIntraPredModeA == candIntraPredModeB and candIntraPredModeA >= 2,

+ if candIntraPredModeA == 2, candModeList[1] and candModeList[2] are 3 and 33 + if candIntraPredModeA == 34, candModeList[1] and candModeList[2] are 2 and 33

I am wondering in these two cases, instead of choosing the opposite direction (33 and 2 respectively), whether neighbouring modes (4 and 32) will be better. In other words,

+ if candIntraPredModeA == 2, candModeList[1] and candModeList[2] are 3 and 4 + if candIntraPredModeA == 34, candModeList[1] and candModeList[2] are 32 and 33

Best regards, Phuong Nguyen.

Hi,

actually, there is the following loops

pbOffset = ( PartMode = = PART_NxN ) ? ( nCbS / 2 ) : 0 for( j = 0; j <= pbOffset; j = j + pbOffset ) for( i = 0; i <= pbOffset; i = i + pbOffset ) { ...

This will cause an infinite loop when PartMode == PART_2Nx2N : for j = 0; j <= 0; j = j + 0.

I think the text should be :

pbOffset = ( PartMode = = PART_NxN ) ? ( nCbS / 2 ) : nCbS for( j = 0; j < pbOffset; j = j + pbOffset ) for( i = 0; i < pbOffset; i = i + pbOffset ) { ...

Best regards, Phuong Nguyen.

Some syntax elements have bypass bins which are not of a prefix and suffix. The value of maxBinIdxCtx specified in table 9-32 might be wrong for some of them.

+ sao_type_idx_luma and sao_type_idx_chroma : maxBinIdxCtx should be 1 instead of 0

+ merge_idx : maxBinIdxCtx can be 1, 2 or 3 but not 0. Since table 9-37 specifies the bypass type for bins 1, 2 and 3, a coherent choice is 3 for maxBinIdxCtx.

Also, in table 9-37, we should fill binIdx 3 and 4 with na.

Best regards, Phuong Nguyen.