Abstract not available in document
JVET-M0902 BoG report on contributions related to complexity analysis and reduction [B. Bross, A. Filippov]
This BoG report was reviewed in Track A on Wednesday 16 January 1515-1600 (chaired by GJS).
There are 10 technical contributions in the complexity related category. These contributions are classified into the following 4 categories:
- Small block size restrictions
- Complexity analysis
- Block size restrictions for PDPC
- Complexity reduction of motion compensation and MV coding
BoG sessions were held 19:15–21:00 on 01/15 in Coliseum. Section 1 of this document summarizes the BoG’s recommendations. Section 2 of this document contains detailed notes on BoG discussion.
The BoG recommended software adoption of JVET-M0864, which suggests an enhancement of the tool to measure the memory bandwidth in VTM.
Decision (SW): Adopt JVET-M0864 memory bandwidth analysis method.
The BoG recommended further discussion of the following:
Summary of contributions on small block size restrictions
Contribution | Coding performance (Y, Cb, Cr) | Disallowed block sizes | Intra prediction loop (throughput) | Other |
(can help both single and separate tree) | AI: 0.04%, 0.23%, 0.31% RA: 0.07%, 0.43%, 0.54% LD: 0.07%, 0.38%, 0.27% | None | 16 sample granularity | Unchanged |
(does not help single tree case) | AI: 0.05%, 0.29%, 0.40% RA: 0.02%, 0.39%, 0.42% LD: -0.01%, 0.33%, 0.50% | Separate tree: Chroma 2x2/2x4/4x2 | 16 sample granularity in separate tree case, 4 otherwise. | Coef. Group: 2x8/8x2 for width/height 2 chroma blocks. |
(not helpful for single tree case in hardware) | Option 1 AI: 0.00%, 0.04%, 0.03% RA: 0.01%, 0.17%, 0.17% LD: 0.03%, 0.27%, -0.11% Option 2 AI: 0.03%, 0.26%, 0.38% RA: 0.09%, 0.63%, 0.84% LD: 0.00%, 0.24%, -0.43 | Option 1: Separate tree: chroma 2x2 Option 2: Separate tree: chroma 2x2/2x4/4x2 | Option 1: 8 sample granularity in separate tree case, 4 otherwise. Option 2: 16 sample granularity in separate tree case, 4 otherwise. | Option 1: Single tree: 2x2 intra use DC only. Option 2: Single tree: 2x2/2x4/4x2 intra use DC only. |
VTM-3.0 without small intra CU | AI: 0.03%, 0.26%, 0.38% RA: 0.33%, 0.55%, 0.78% LD: 0.21%, 0.17%, 0.60% | Chroma: 2x2,4x2,2x4 Luma: 4x4,4x8,8x4 | 16 sample granularity |
It was commented that it may be undesirable to depend on requiring decoders to be able to use parallel processing, since some (e.g., software) decoders may not be able to do that.
It was agreed to consider these approaches in CE3.
Summary of contributions on block size restrictions for PDPC
Contribution | Test # | Restriction | Compression performance (Y/U/V) |
width + height <= 8 | AI: 0.05% / 0.02% / 0.03% RA: 0.03% / 0.04% / 0.00% | ||
1 |
| AI: 0.03% / 0.04% / 0.02% RA: 0.03% / 0.03% / 0.07% | |
2 |
| AI: 0.03% / 0.03% / 0.04% RA: 0.02% / 0.06% / 0.05% | |
3 |
Both restrictions are applied only to blocks with worst-case intra-prediction modes | AI: 0.01% / -0.03% / 0.00% RA: 0.02% / 0.07% / 0.04% | |
3a | width + height <= 8 (applied only to blocks with worst-case intra-prediction modes) | AI: 0.02% / -0.03% / -0.07% RA: 0.01% / 0.01% / -0.04% | |
width + height > 64 | AI: -0.01% / -0.02% / 0.02% RA: 0.00% / 0.08% / 0.10% | ||
1 | width * height <= 16 | AI: 0.05% / 0.01% / 0.03% RA: 0.03% / -0.05% / -0.01% | |
2 |
| AI: 0.06% / -0.08% / -0.08% RA: 0.05% / -0.08% / -0.05% | |
3 | width * height <= 8 for chroma | AI: 0.00% / 0.03% / 0.02% RA: 0.00% / 0.06% /0.02% |
It seemed unclear whether PDPC is particularly an issue for small blocks.
A participant indicated that it may be too early in the process, so that we shouldn’t optimize such aspects until the design is more stable.
No action was taken on these proposals, and no CE for them was planned.
M0248 was reviewed in the context of CE2 rather than in this BoG.
M0265 was not reviewed in the BoG. It was considered in Track A, and notes for that are recorded separately.
JVET-M0902 BoG report on contributions related to complexity analysis and reduction [B. Bross, A. Filippov]
See section 7.