JVET-R0129 AHG9/AHG12: On CTU row based slice chunks of a slice within a tile [L. Chen, C.-W. Hsu, C.-C. Chen, Y.-L. Hsiao, C.-Y. Chen, T.-D. Chuang, C.-M. Tsai, Y.-W. Huang, S.-M. Lei (MediaTek)]
Discussed in AHG Session 1.10 Wednesday 8 April at 1630 UTC (chaired by GJS & YKW).
This contribution proposes to specify CTU row-based slice chunks of a slice within a tile in decoding order. The slice chunks in sequence of a slice are delivered and decoded sequentially. Each slice chunk is proposed to be contained in a single NAL unit. It is claimed that no essential new requirements are needed in the decoding process. It is claimed that the proposed slice chunks can meet ultra-low latency requirements with better coding efficiency than conventional slices. It is reported the BD-rate savings of the proposed slice chunks compared against the conventional slices are 4.52%, 3.55%, 3.64%, and 4.93% for short length (1 sec) Class A1 (4K), Class A2 (4K), Class B (1080p), and Class E (720p), respectively, when the encoding latency is one CTU row. Proposed reference software was claimed to be ready and cross-checked and was attached in the uploaded proposal package.
This would be something like a special case of the dependent slice segment concept of HEVC. The proponent said dependent slice segments was a useful feature of HEVC, while another participant said they had seldom encountered this feature in practice.
Especially given the late stage at which we are in this design process, there were strong objections to the proposed change, so no action was taken on this.