Organisation internationale de normalisation

JCTVC-U0177 Crosscheck for the Joint proposal of JCTVC-U0169 and JCTVC-U0090 Item 3 (JCTVC-U0176) [W. Zhang (Intel)] [late]

The cross-checker verbally confirmed most AI results; others were ongoing.

2. CE2 related (intra block copy memory access) (1)

(Consideration of this topic was chaired by JRO on Friday 06-19, 16:15-16:30.)

1.1.1.1.1.1.1.1.106JCTVC-U0065 CE2-related: Intra block copy searching constraints for reducing worst case bandwidth [T.-D. Chuang, C.-Y. Chen, Y.-W. Huang, S. Lei (MediaTek)]

In HEVC-SCC, the IntraBC predictor comes from the unfiltered reconstructed samples of the current picture. The unfiltered reconstructed samples need to be accessed, which increases the worst-case memory bandwidth for HEVC-SCC. In this contribution, the worst case bandwidth of HEVC and HEVC-SCC is analyzed. Two IntraBC encoder searching constraints are proposed to reduce the worst case bandwidth.

In Method-1, a current picture is divided into 8×8 blocks. An 8×8 block cannot be referenced by IntraBC blocks if this 8×8 block is coded as an 8×8 bi-predicted block, two 4×8/8×4 uni-predicted blocks, or it belongs to a 16×16 CU that is coded by 4×16+12×16 bi-predicted blocks or 16×4+16×12 bi-predicted blocks, and the 8×8 block is not an IntraBC coded block and the use_integer_mv_flag is 0. In experiment results for method-1, no BD-rate losses are reported under AI configuration, 0.5% to 0.6% BD-rate losses are shown under RA configuration, and 0.7% to 1.0% BD-rate losses are reported under LB configuration.

In method-2, an additional constraint is added that if one MV in this 8×8 block is integer MV, this 8×8 block is allowed to be referenced by IntraBC blocks. In experiment results for method-2, no BD-rate losses are reported under AI configuration, 0.3% to 0.4% BD-rate losses are shown under RA configuration, and 0.2% to 0.6% BD-rate losses are reported under LB configuration.

With the proposed methods, the worst case bandwidth of the HEVC-SCC reportedly will be the same as the worst-case bandwidth of the current HEVC design.

This is a similar idea as in CE2 test 5, but this is in the form of an encoder/bitstream constraint.

The contribution did not contain draft text, but it was asserted that it would not be simple to specify the method as a bitstream constraint, and may not be easy for an encoder to implement it.

No action was taken on this.

3.