Organisation internationale de normalisation
SCC tool complexity (AHG9) (0)
Yüklə 5,54 Mb.
|
SCC tool complexity (AHG9) (0)No contributions specific to this topic were noted, although various contributions included consideration of complexity issues.
1.1.1.1.1.1.1.1.140JCTVC-U0106 On signalling adaptive colour transform at the TU level [X. Xiu, Y. He, Y. Ye (InterDigital)] (Consideration of this topic was chaired by GJS on Sunday 06-21, 10:15-10:45.) In HEVC screen content coding draft 3, the adaptive colour transform is enabled/disabled at the CU level such that all the residuals of the CU are coded in the same colour space. This contribution proposes to improve the performance of adaptive colour transform by moving the adaptive colour transform flag to the TU level. Compared to the SCM-4.0 anchor using full-frame IBC search, the proposed solution reportedly provides average {G, B, R} BD-rate reductions of {0.2%, 0.5%, 0.4%}, {0.1%, 0.6%, 0.5%} and {0.3%, 0.4%, 0.2%} for AI, RA and LD, respectively, in RGB coding, and average {Y, Cb, Cr} BD-rate reductions of {0.6%, 2.0%, 1.9%}, {0.3%, 1.4%, 1.4%} and {0.1%, 0.3%, 0.3%} for AI, RA and LD, respectively, in YCbCr coding. For lossless coding, the corresponding bit-rate reductions are reportedly 0.4%, 1.1% and 1.0% for AI, RA and LD in RGB coding, and 0.0%, 0.0% and 0.0% for AI, RA and LD in YCbCr coding. CCP is currently controlled at the TU level, whereas ACT is controlled at the CU level (which seems potentially inconsistent). Some encoder speed-up techniques were also included in the contribution. It was remarked that some of the gain for YUV intra cases may be associated with performing a joint optimization of intra prediction modes for three colour components together rather than performing them separately, which is something that could be done without the normative change. Approximately a 5% run-time increase was reported for all-intra coding. The cross-checker reported a larger run-time increase, but the cross-checker said that the run-times of the cross-check were likely not to reliable due to heterogeneous computing cluster issues. The motivation for the proposal is partly harmonization with the existing CCP operation, not just coding efficiency improvement. Decision (cleanup): Adopt (editorially, it was suggested to consider naming the flag to tu_residual_act_flag). 1.1.1.1.1.1.1.1.141JCTVC-U0165 Cross-check of on signalling adaptive colour transform at TU level (JCTVC-U0106) [B. Li, J. Xu (Microsoft)] [late]
(Consideration of this topic was chaired by GJS on Sunday 06-21, 15:40-14:10.) This document proposes a segment freeze process using high-level syntax elements. It was stated that the principle of the proposed freeze process consists of the freezing of a slice segment in the current picture for subsequently coded pictures. It was proposed that no syntax be coded in these subsequent pictures to reconstruct the samples of the part of the picture corresponding to the frozen slice segment. Instead, it was proposed that the frozen slice segment be copied to the corresponding region of these subsequent pictures. Additionally, a termination method for the freeze process was proposed to end the freezing of a slice segment. For "frozen" slices, there would be a picture containing a coding of the slice area and a flag indicating that the area will then be "frozen". In subsequent pictures, missing areas will be filled in by the decoder. It was asked what picture would be used to fill in the missing regions – e.g., a particular picture in the DPB or just the previous picture in decoding order, and whether extra memory would be required to retain a copy of this data. Some somewhat similar proposals had previously been submitted. It was commented that using low-level syntax instead of high-level syntax to accomplish the goal would not really have much overhead. It was asked what happens if the slice that was intended to "unfreeze" a region is lost due to packet loss. The loss of the "unfreeze" might not be detectable. It was commented that, in some sense, the functionality could be accomplished without the proposed flag. The flag would be unnecessary if there was a "normative error concealment" specified for how to fill in missing areas of a picture using some reference picture. The proposed flag is just a warning that this is happening intentionally. No action was taken on this. 1.1.1.1.1.1.1.1.143JCTVC-U0138 Chroma Deblocking for Screen Content Coding [A. M. Tourapis, Y. Su, D. Singer (Apple), W. S. Kim, W. Pu, J. Sole, M. Karczewicz (Qualcomm)] [late] (Consideration of this topic was chaired by GJS on Sunday 06-21, 16:10-16:45.) This contribution proposes extending the deblocking process for any new screen content coding profile with two additional deblocking modes. In the first additional proposed mode, the deblocking of chroma edges is extended to also cover edges with a boundary strength equal to 1. In the second proposed mode, which only applies to 4:4:4 encoding, chroma deblocking is performed using the luma deblocking filter process. These extensions were described as being motivated primarily from additional observations on the quality of the chroma components after HEVC decoding, as well as due to the unification of the intra block copy mode with inter prediction. The selection of deblocking operation was proposed to be indicated at the PPS level. It was suggested that the issue this is trying to address is not specific to SCC (other than the interaction with IBC), but is rather intended to be useful for all content. It was remarked that similar proposals had previously been considered in earlier work on version 1 and RExt, but that efforts to prove there was a significant benefit to changing the DBF had not succeeded previously. The contributor reported that recent work on HDR coding had shown a benefit for modified deblocking (as a post-filter, which would likely have performed better as a loop filter), both in objective and subjective terms. Tests on SCC content (e.g., SCC CTC content), had not been performed. Text was not provided, but probably would not be very substantial. Text for something similar had been provided previously in JCTVC-O0089. It was suggested to perform a CE to study this.
|