International organisation for standardisation organisation internationale de normalisation
Yüklə 8,24 Mb.
|
5.16Transform coefficient coding5.16.1Significance map codingAll contributions in this category were reviewed in a BoG (H0719) coordinated by G. Martin-Cocher. 5.16.1.1.1.1.1.1.1JCTVC-H0719 BoG report on significance map [G. Martin-Cocher] The break out group reviewed contributions from CE11, and also non-CE11 contributions related to significance map and level coding. Contributions were classified as multilevel significance map contributions; simplification and harmonization of 4x4 and 8x8 significance map contributions; and parallelization and context template optimization contributions. Recommendations from the BoG were as follows:
Further remarks from the BoG: H0100 (first part of the proposal) is to have harmonization between 4x4 and 8x8 luma + 4x4 and 8x8 chroma. 11 context reduction, one look-up table less. H0282: no context map. 7 more contexts. No loss in normal QP, up to 0.1% loss for inter case. Can be implemented with simple logic, without look-up tables. Was studied in CE11. H0284: one context map for 4x4 and 8x8. Option 1 keeps 4x4 chroma. Option 2: keep 4x4 luma. No losses by down-sampling. DC is different. One expert expressed support for H0282, as a good trade-off between complexity and additional contexts. One expert indicated that both context reduction and simple logic would be desirable, which is not achieved in H0282. Another expert expressed that context reduction is still meaningful and H0100 proposes to reduce the number of contexts by around 10%. Some experts expressed support for H0284. One expert expressed that having DC as a different step as in H0284 is an extra complexity. Further study (in an AHG on harmonization of TU size independent coefficient coding) was suggested. This also includes H0379 and H0399 “item 3” and SQT/NSQT harmonization as previously tried in CE11 subtest 4. Co-Chairs for this work were suggested to include V. Sze, G. Martin-Cocher, and J. Sole. 5.16.1.1.1.1.1.1.2JCTVC-H0100 Non-CE11.1: Context reduction of significance map coding [T. Tsukuba (Sharp)] 5.16.1.1.1.1.1.1.3JCTVC-H0620 Non-CE11: Cross-check report [H. Sasai, K. Terada, T. Nishi (Panasonic)] [late] 5.16.1.1.1.1.1.1.4JCTVC-H0131 Non-CE11: Simplified multi-level significance map coding [J. Min, Y. Piao, J. H. Park (Samsung)] This proposal removes one check performed during the multi-level significance map coding (condition removed: if lower and right significant_coeffgroup_flag in the neighbouring groups are both 1, significant_coeffgroup_flag is also inferred as 1). No losses were reported, with some reduction in number of bins to be coded. Decision: Adoption of JCTVC-H0131. 5.16.1.1.1.1.1.1.5JCTVC-H0488 Cross-check of Samsung H0131: Simplified multi-level significance map coding [N. Nguyen, T. Ji, D. He, G. Martin-Cocher (RIM)] [late] 5.16.1.1.1.1.1.1.6JCTVC-H0253 Improvement on parallelism for context derivation process of significant_coeff_flag [A. Minezawa, K. Sugimoto, S. Sekiguchi (Mitsubishi)] This contribution is the same as H0427 method 2, which is planned to be studied in CE work. 5.16.1.1.1.1.1.1.7JCTVC-H0599 Cross-verification of Mitsubishi improvement on parallelism for context derivation process of significant_coeff_flag (JCTVC-H0253) [K. Chono, H. Aoki (NEC)] [late]
Decision: Adopt JCVTC-H0290 second pattern (both for 16x16 and 32x32). 5.16.1.1.1.1.1.1.15JCTVC-H0357 Non-CE11: Cross-check of proposal H0290 on significant_coeff_flag context selection [V. Sze (TI)] [late 02-02]
This proposal is to apply the multilevel significance map to 8x8 (vertical and horizontal) by reusing the same scans of MDCS. The number of bins saved is up to 1.9%. No losses were reported. Inferred flags don’t depend anymore on the right and lower neighbours, which was anyway removed by JCTVC-H0131. Some experts suggested that this is a good simplification/unification. There was a request to review the working draft text, and it was noted that the semantics text needs to be improved – and established that this would be done during the meeting. Experts expressed support for method B. Decision: Adoption of JCTVC-H0526 method B (working draft text of H0526r2 was reviewed by B. Bross and code was reviewed by various people in the breakout). 5.16.1.1.1.1.1.1.32JCTVC-H0615 Cross-check of JCTVC-H0526 on multi-level significance maps for small TU [Y. Piao, E. Alshina, J. H. Park (Samsung)] [late] 5.16.1.1.1.1.1.1.33JCTVC-H0547 Significance map contexts reduction for small TU [V. Kung, K. Panusopone, L. Wang (Motorola Mobility)] 5.16.1.1.1.1.1.1.34JCTVC-H0628 Cross-check of Motorola proposal on significance map contexts reduction for small TU (JCTVC-H0547) [H. Yang, H. Yu (Huawei)] [late 02-04] 5.16.1.1.1.1.1.1.35JCTVC-H0678 Non-CE11: Signalling of significant_coeffgroup_flag for significance map [V. Sze, A. Osamoto (TI)] [late] Yüklə 8,24 Mb. Dostları ilə paylaş:
|