spec, software mismatch of boundary partition for corner case

[hangao]

For example, if a 64x64 CU located on the corner (both bottom-left sample and top-right sample are outside of the picture boundaries) and CU size is greater than MinQtSize case

In the software: only QT is allowed, no signaling is needed.

In the spec: split_cu_flag is inferred to be 1, however, split_qt_flag is signaled because both allowSplitQT and allowSplitBtHor are true based on the following table in JVET-N1001-v2 page 42

if( split_cu_flag ) {

if( ( allowSplitBtVer | | allowSplitBtHor | | allowSplitTtVer | | allowSplitTtHor ) && allowSplitQT )

split_qt_flag

A fix could be as following by adding the bold part into 6.4.2:

6.4.2 Allowed binary split process
…
– Otherwise, if all of the following conditions are true, allowBtSplit is set equal to FALSE

– btSplit is equal to SPLIT_BT_HOR
– cbWidth is greater than MaxTbSizeY
– y0 + cbHeight is greater than pic_height_in_luma_samples

– Otherwise, if all of the following conditions are true, allowBtSplit is set equal to FALSE

– x0 + cbWidth is greater than pic_width_in_luma_samples
– y0 + cbHeight is greater than pic_height_in_luma_samples
– cbWidth is greater than minQTSize

– Otherwise, if all of the following conditions are true, allowBtSplit is set equal to FALSE

– btSplit is equal to SPLIT_BT_HOR
– x0 + cbWidth is greater than pic_width_in_luma_samples
– y0 + cbHeight is less than or equal to pic_height_in_luma_samples

…

Thanks Adam for the crosscheck

[bbross]

Fixed in D5 v10