International organisation for standardisation
Yüklə 9,08 Mb.
|
5.11.2Contributions5.11.2.1.1.1.1.1.1JCTVC-F128 CE11: Reduced neighboring dependency in context selection of significant_coeff_flag for parallel processing (JCTVC-E330) [V. Sze, M. Budagavi (TI)] This contribution (JCTVC-E330 in Geneva) proposes reducing the neighboring dependencies in the context selection of significant_coeff_flag for positions in the top-most row and left-most column of a transform (i.e. the top and left edge). While the simplified context selection method in JCTVC-D260 (adopted for HM-2.0), eliminates the neighboring dependency on the most recently decoded significant_coeff_flag within a diagonal, this contribution eliminates the dependency when wrapping from one diagonal to another during zigzag scan. This modification was implemented in HM-3.0 and its coding efficiency was evaluated in the high efficiency configuration with a coding loss 0.1%, 0.1%, 0.0% for Intra and Random Access and Low Delay. The proposal is to omit a neighbor at the turning points of the diagonal scan. BR is increased by 0.1% 5.11.2.1.1.1.1.1.2JCTVC-F185 CE11: Cross-check result of TI proposal on context simplification of the significance map (JCTVC-F128). [C. Rosewarne, M. Maeda (Canon)] 5.11.2.1.1.1.1.1.3JCTVC-F311 CE11.A: Cross checking of TI proposal JCTVC-E330 on context simplification of significance map [C. Auyeung (Sony)] 5.11.2.1.1.1.1.1.4JCTVC-F129 CE11: Parallelization of HHI_TRANSFORM_CODING (Fixed Diagonal Scan) [V. Sze, M. Budagavi (TI)] This contribution proposes the use of a fixed diagonal scan (JCTVC-C227) for the significance map to enabled parallel processing in conjunction with the use of the highly adaptive context selection for significant_coeff_flag. While the simplified context selection method in JCTVC-D260 (adopted for HM-2.0), eliminates the neighboring dependency on the most recently decoded significant_coeff_flag within a diagonal, dependencies on the previous position when moving from one scan line to another. The fixed diagonal scan proposed in this contribution eliminates this dependency. The fixed diagonal scan is only proposed when CABAC is used (i.e. HE configuration). Note: JCTVC-F134 provides results on supporting fixed diagonal scan for CABAC described in this contribution with the existing zigzag scan for CAVLC. The fixed diagonal scan was implemented in HM-3.0 and had coding efficiency impact of −0.1%, −0.1% and 0.0% for All Intra, Random Access and Low Delay for the High Efficiency configuration. The proponents would recommend using this method (with bottom-left-to-top-right scan) rather than JCTVC-F128 as it does not require specific context switching and has slightly better performance. Decision: Adopt fixed diagonal scan of significance map (bottom-up) as suggested in JCTVC-F129. 5.11.2.1.1.1.1.1.5JCTVC-F293 CE11: Cross-check of TI’s fixed diagonal scan (JCTVC-F129) [J. Sole (Qualcomm)]
In JCTVC-E335, a unification of scans for significance map coding and coefficient level coding in high efficiency was proposed. This contribution analyses and shows the results of the technique in HM3.0. It consists of two parts. In the first one, the scan for the coefficient level coding is the same as the scan used for the coding of the significance map, but in reverse order, i.e., from the last significant coefficient to the first (DC). For this part, the BD-rate for AI-HE, RA-HE and LB-HE configuration is −0.05%, 0.07%, and −0.01%, respectively. In the second part, scans for significance map coding and coefficient level coding are both in the reverse order. For this case, the BD-rate for AI-HE, RA-HE, and LB-HE is −0.25%, −0.19%, and −0.07%, respectively. Additional result is shown about combination with the diagonal scan of JCTVC-F129. This has not been cross-checked Question: A new context initialization was used which may be the main reason for the gain in part 2 and no loss in case 1. Was this trained towards the test set? Original contributors of the two different scans argue that usage was shown to be advantageous earlier, and potentially with tuned context initialization could reproduce such gain again. It is also mentioned that JCTVC-F124 and JCTVC-F569 may be related. On request, JCTVC-F597 was also discussed in this context. It was argued that by re-ordering, a similar advantage in terms of computation time could be achieved as in JCTVC-F288. Another expert raises the opinion that a single scan (without need of reordering) would be beneficial for software implementation. Breakout group (R. Joshi) will try to clarify this issue, and report back when other transform coefficient coding is discussed. 5.11.2.1.1.1.1.1.8JCTVC-F136 CE11: Cross-check of Qualcomm's Unified scans for the significance map and coefficient level coding in high coding efficiency (JCTVC-E335 Part 2) [V. Sze (TI)] (Another cross-check is in JCTVC-F134.) 5.11.2.1.1.1.1.1.9JCTVC-F369 CE11:Cross-check report for Qualcomm's proposal JCTVC-F288 part2 [H. Sasai, T. Nishi (Panasonic)] 5.11.2.1.1.1.1.1.10JCTVC-F451 CE11: Crosscheck - Reduced neighboring dependency in context selection of significant_coeff_flag for parallel processing (JCTVC-E330/JCTVC-F128) [T. Nguyen] 5.11.2.1.1.1.1.1.11JCTVC-F597 CE11: Hardware complexity of large zigzag scan for level-coding of transform coefficients [C. Auyeung, T. Suzuki (Sony)] See notes above in discussion of JCTVC-F288. 5.11.2.1.1.1.1.1.12JCTVC-F607 CE11:Cross-check report for Qualcomm's proposal JCTVC-F288 part1 [J. Chen, V. Seregin (Samsung)]
Yüklə 9,08 Mb. Dostları ilə paylaş:
|