International organisation for standardisation organisation internationale de normalisation
Yüklə 8,24 Mb.
|
4.10.2Contributions4.10.2.1.1.1.1.1.1Subtest 1 4.10.2.1.1.1.1.1.2JCTVC-H0084 CE10 Subtest 1: Test results on unification of luma and chroma filtering [M. Zhou (TI)] 4.10.2.1.1.1.1.1.3JCTVC-H0147 CE10.1: Symmetric rounding in the deblocking filter [M. Narroschke, A. Kotra, T. Wedi (Panasonic)] The current HM-5.0 deblocking filter generates an offset due to a non-symmetric rounding operation. Proposed is a symmetric rounding operation to eliminate this undesired offset. Compared to HM5.0, the following BD-bit rate reductions are achieved at similar complexity: I-HE: -0.1%, I-LC: -0.1%, RA-HE: -0.1%, RA-LC: 0.0%, RA-HE-10: 0.0%, LD-HE(B): -0.1%, LD-LC(B): -0.1%, LD-HE(P): 0.0%, LD-LC(P): 0.1%. Measured run-time ratios are all around 100%. It was proposed to adopt the symmetric rounding into the next version of the HM. It was remarked that the domain in which this asymmetric rounding occurs might be in a position where the asymmetry doesn't matter. It was remarked that the proposed modification seems to increase complexity, depending on how it is implemented. 4.10.2.1.1.1.1.1.4JCTVC-H0262 CE10 Subtest1: Cross-verification for Panasonic proposal JCTVC-H0147 [M. Ikeda, T. Suzuki (Sony)]
4.10.2.1.1.1.1.1.11JCTVC-H0149 CE10.2: Cross check results for the deblocking filter line buffer reduction of Sony and MediaTek [A. Kotra, S. Esenlik, M. Narroschke (Panasonic)] 4.10.2.1.1.1.1.1.12JCTVC-H0264 CE10 Subtest2: Cross-verification for Ericsson proposal [M. Ikeda, T. Suzuki (Sony)] 4.10.2.1.1.1.1.1.13JCTVC-H0280 CE10.2.1: Reducing one pixel line buffer by modified deblocking filter for horizontal LCU boundaries [C.-W. Hsu, C.-Y. Chen, Y.-W. Huang, S. Lei (MediaTek), M. Ikeda, T. Suzuki (Sony)] This contribution proposed to modify the deblocking in HM-5.0. The luma vertical filtering for the upper side of the horizontal LCU boundary reads three pixels and writes two pixels (instead of four and three), and the other parts of deblocking are not changed. The change is made only for LCU boundaries. The proposed method can save one pixel line buffer. It is reported that no changes in BD bit rates, run times, and subjective quality are observed. Moreover, the luma virtual boundary of adaptive loop filtering is shifted to three pixels above the horizontal LCU boundary to match the proposed deblocking. Simulation results also reportedly show no changes in BD bit rates, run times, and subjective quality. It was remarked that we should be very conservative about line buffer reduction attempts, in fear of line-structured artefacts, and that line buffers are not so much of a terrible burden (e.g. they would be present in an AVC decoder, and most HEVC decoders would probably also be AVC decoders). It was remarked that the "slideshow" sequence has loss in multiple conditions, e.g. LP and LB. No action taken. 4.10.2.1.1.1.1.1.14JCTVC-H0472 CE10.2: Level restrictions on tiles for line buffer reduction [A. Norkin, R. Sjöberg (Ericsson)] No apparent visual impact. This should be considered in profile/level definition work. 4.10.2.1.1.1.1.1.15JCTVC-H0575 CE10.2: cross-verification of Sony and Mediatec proposal on deblocking line memory reduction [A. Norkin (Ericsson)] [late]
4.10.2.1.1.1.1.1.17JCTVC-H0108 CE10 Subtest3: Simplified BS calculation process in deblocking filter [S. Park, B. Jeon (LG)] Part 1 is discussed below. Part 2 reduces BS calculation complexity by using 4 samples to calculate the BS and using that value for the next 4 samples as well. No action was taken on this aspect. 4.10.2.1.1.1.1.1.18JCTVC-H0146 CE10.3: Simplified delta calculation in the deblocking filter [M. Narroschke, A. Kotra, T. Wedi (Panasonic)]
The contribution proposed a simplification of the deblocking filter by simpler calculation of the second delta value. Simulation results reportedly show that the BD bit rate results and subjective quality are equivalent to the HM5.0 anchor with a benefit of reduced decoding execution time. It was remarked that this could amplify high frequencies, as was previously remarked in Turin. It was remarked that there are significant losses for LC LP in some cases. 4.10.2.1.1.1.1.1.21JCTVC-H0157 CE10.3: Cross verification of SKT/SKKU Simplification of Second Delta Calculation in Deblocking (JCTVC-H0156) [S. Lee, N. Eum (ETRI)] [late]
This contribution contains two aspects, part 1 of which is within the scope of CE10 and part 2 of which was new. Part 2 should be considered to be outside of CE10. Part 1 modifies the delta calculation to simplify it for luma and make the way it is done for chroma consistent with the way it is done for luma. The modification for luma makes the effect of the filter weaker. There does not seem to be a clear case for doing that. No action was taken on this. 4.10.2.1.1.1.1.1.24JCTVC-H0209 CE10.3: Cross-verification Results of Qualcomm Simplification of Deblocking (JCTVC-H0396) by SKT/SKKU [J. Yang, J. Kim, K. Won, B. Jeon (SKKU), J. Lim (SKT)] [late] 4.10.2.1.1.1.1.1.25JCTVC-H0459 CE10.3: Cross Check of SKT/SKKU Proposal JCTVC-H0155 [G. Van der Auwera (Qualcomm)] 4.10.2.1.1.1.1.1.26JCTVC-H0473 CE10.3: Deblocking filter simplifications: Bs computation and strong filtering decision [A. Norkin (Ericsson)] Part 1 removes a loop on 4-pixel boundaries inside of 8-pixel boundaries. The current structure seems a bit obtuse. Decision: Adopt part 1. (This is also part 1 of H0108.) Part 2 is a simplification of the strong filter decision. No action was taken on that part. 4.10.2.1.1.1.1.1.27JCTVC-H0618 Cross-verification of Ericsson proposal on deblocking filter simplifications (JCTVC-H0473) [S. Park, B. Jeon (LG)] [late]
4.10.2.1.1.1.1.1.30JCTVC-H0148 CE10.4: Deblocking filter using adaptive weighting factors [A.-K. Seifert, T. Wedi, M. Narroschke (Panasonic)] 4.10.2.1.1.1.1.1.31JCTVC-H0350 CE10 Subtest4: Cross-verification for Panasonic proposal JCTVC-H0148 [M. Ikeda, T. Suzuki (Sony)] [late] 4.10.2.1.1.1.1.1.32JCTVC-H0398 CE10 Subtest 4: Transform Dependent Deblocking Filter Parameter Adjustment [G. Van der Auwera, R. Joshi, X. Wang, M. Karczewicz (Qualcomm)] One aspect of the proposal is an SPS-level flag to save one bit per slice (assuming that the semantics of inherit_dbl_params_from_APS_flag = 1 will result in inferred values of zero in the case where no parameters have been provided in the APS) when the flag is set to indicate that deblocking control parameters are never sent. In AVC, there was a conceptually-similar flag at the PPS level. This is actually a high-level syntax issue rather than a deblocking filter design issue. Decision: Put such a flag in the PPS level as in AVC, rather than in SPS, as proposed. 4.10.2.1.1.1.1.1.33JCTVC-H0656 CE10 Subtest 4: Cross-check of Qualcomm Transform-Dependent Deblocking Filter Parameter Adjustment JCTVC-H0398 [T. Yamakage, S. Asaka (Toshiba)] [late] 4.10.2.1.1.1.1.1.34JCTVC-H0474 CE10.4: Cross check of Ericsson proposal [G. Van der Auwera (Qualcomm)] [late] 4.10.2.1.1.1.1.1.35JCTVC-H0574 CE10.4: deblocking parameters signalling [A. Norkin, K. Andersson, J. Enhorn, R. Sjöberg (Ericsson)] [late] Yüklə 8,24 Mb. Dostları ilə paylaş:
|