Joint Video Experts Team (jvet) of itu-t sg 6 wp and iso/iec jtc 1/sc 29/wg 11
Yüklə 4,04 Mb.
|
- Bu səhifədəki naviqasiya:
- 7.19Palette mode (2)
- [8]Complexity analysis and reduction (8)
7.18Resilient intra refresh (12)Contributions in this category were discussed Tuesday 17 July in Track B 0930–XXXX (chaired by JRO). JVET-K0212 Improved Cyclic Intra Refresh [J.-M. Thiesse, D. . Nicholson] [late] This contribution proposes a technique relating to improve the use of cCyclic iIntra rRefresh, a technique that is reportedly commonly used for real-time implementation of HEVC video encoders. This cyclic iIntra Rrefresh reportedly provides several desirable features, such as bit-rate smoothing, maintaining keeping iIntra data bit-rate under control, and enabling low delay and natural error resilience while limiting temporal error propagation along frames. For The contributor said that for video encoder’s manufacturers this feature is considered as mandatory for real-life applications. In VVC rRequirements documents, as contained in VCEG-BD03 and WG11/ N17074, low latency and error resilience requirements are present but were said not to have not been addressed by the responses to the CfP. One big factor of latency apart from picture ordering and related dependencies, is the importance of iIntra picture data amongst other issues, with an increasing iIntra picture data size versus iInter picture data size ratio at for each video codec generation of video coding design. Cyclic iIntra rRefresh can be usedwas said to be appropriate for fulfilling these requirements. But However, this feature, as applied by encoder manufacturer in for use with previous standards, implies some encoding restrictions –, limiting the coding efficiency, as well as rRandom aAccess capability. The proposed technique through appropriate signalling enablesis claimed to leverage these limitations, allowing less coding restrictions. In this document, experiment results were provided that were asserted to show an improvement toward classic use of cCyclic iIntra rRefresh together with the usual LD configuration as a reference. The aAuthors recommended considering the proposed technique for a potential inclusion in the VVC future standard, with further study to be conducted in an appropriate Core ExperimentCE. Some support is iswas expressed for studying the aspect of periodic intra refresh, which is commonly used in practice, but although it was said that there are more approaches for resolving it. The bBenefit of vertical refresh regions was said to be not that obvious (except for ultra low delay, as it has less bit rate fluctuation over the frame). Horizontal refresh could be done with slices. However, independent slices cause some bit rate overhead. A nNon-normative solution approach (just enforcing intra locally) causes up to 25% BR loss (the maximum for class E). By introducing normative signalling of the vertical refresh area, 2.5% is reportedly recovered. For other classes, the loss of the non-normative solution is less than <10%, of which approx.imately 3.5% is recovered. These results, however, reflect the case where the whole sequence is decoded from the beginning. As periodic intra refresh is used for random access, it wshould better be tested better to test what happens if the decoding of a bitstreams starts at some random picture position. Also, the SPS?/PPS and possibly SPS should be transmitted with every picture and counted in the bit rate, as well asand the propagation errors should be counted in terms of quality. Further study was encouraged (in an AHG on low latency coding, coordinated by J.-M. Thiesse, A. Duenas, and A. Tourapis).
X-check
Contributions in this category were discussed Tuesday 17 July in Track B 0900–0930 (chaired by JRO). JVET-K0411 AHG11: Palette mode [Y.-C. Sun, J. . An, J. . Lou (Alibaba)] [late] This document proposes a simplified palette mode on top of VTM1.1. For Class F sequences, the results reportedly show -2.6%/-1.8%/-0.9% BD-rate luma gain in AI/RA/LDB configuration; for 4:2:0 TGM sequences in CE8, the results reportedly show -13.7%/-8.9%/-7.0% BD-rate luma gain in AI/RA/LDB configuration. The proposed pPalette mode is not the same as in HEVC, but rather similar to an earlier version (SCM2). This was aApplicable to 4:4:4 and 4:2:0 colour sampling. The mode is eEnabled at the CU level (which is now rectangular, whereas it was always square in the case of HEVC. Small losses were reported for classes A through E. To be iInvestigated in CE15:
JVET-K0475 Crosscheck of JVET-K0411: AHG11: Palette mode [J. . Ye, S. . Liu (Tencent)] [late] [8]Complexity analysis and reduction (8)Contributions in this category were discussed XXday XX July XXXX–XXXX (chaired by XXX). JVET-K0057 Reduced Memory Storage for Collocated Picture [Y. . Yu, S. . Hong, K. . Panusopone, L. . Wang (Arris)] The contribution was available for study, but no presenter was available for the contribution. JVET-K0086 Non-CE: Temporal motion vector data storage reduction [H. . Jang, J. . Lim, J. . Nam, S. . Kim (LGE)] This contribution reports coding performance impact with respect to the compression unit size of temporal motion vector storage base on spatial candidates located in compression unit. It is observed that the proposed compression method shows 0.1% BD-rate coding gain from 16x16 compression unit on VTM Random access configuration. The proposed method achieved that 0.1% in partial result and 0.28% BD-rate loss with a factor of 4 and 16 memory size reduction respectively in BMS Random access configuration. Was reviewed in track B Monday afternoon Selection of the vector for temporal storage is done based on spatial candidate. Results are provided for the cases of 16x16 (as HEVC), 32x32 and 64x64 memory compression. (update table from new version and fix column alignment and margins)
The complexity in number of operations is almost negligible. The gain in VTM (approx. 0.1%) shown for the case of 16x16 indicates that the method has benefit. Further study (CE4) in combination with the new 8x8 compression scheme of VTM. JVET-K0530 Crosscheck of JVET-K0086: Temporal motion vector data storage reduction [Y. . Han, W.-J. Chien (Qualcomm)] [late] JVET-K0106 Energy Assessment for Video Decoding [C. . Herglotz (FAU)] [late] This was discussed Sunday 15 July 1645 (GJS & JRO). This proposal describes a contribution on the assessment of processing energy for video decoding. It is stated that the processing energy is a relevant metric especially for portable devices like smartphones or tablet PCs, because small energy requirements help to increase the operating time until the battery is empty. It is proposed to use energy measurements to assess the energy efficiency of the decoding process. In this proposal, a general method for energy measurements is presented that can reportedly be used for various video coding platforms. The results can be used to evaluate a decoder in terms of processing energy. Finally, it is proposed to use the Bjøntegaard-Delta Metric to assess the energy efficiency of the complete decoding process. Study and experimentation with the method was encouraged.
This was discussed Sunday 15 July 1655 (chaired by GJS & JRO). In this contribution, a recently published energy model is presented which can be used for video decoding energy estimations. It is reported that the model is able to estimate the energy of both the processor and the memory access. Furthermore, it is stated that using sophisticated training methods, the trained model values can be interpreted as coding tool specific energies, and that the model can be used for decoder energy optimization. The model is based on the amount of usage of a technical element and the energy used when that element is exercised. Further study was encouraged. The slide deck was requested to be provided in a revision of the contribution. JVET-K0108 Decoding-Energy-Rate-Distortion Optimization [C. . Herglotz, M. . Kränzler, A. . Kaup (FAU)] [late] This was discussed Sunday 15 July 1505 (chaired by GJS & JRO). This contribution proposes to include “decoding-energy-rate-distortion optimization” (DERDO) into the encoder reference software. It is reported that DERDO is an extension of RDO which can be used to control and minimize the decoding energy on the receiver side which is interesting for portable, battery driven devices. As the complexity of the next coding standard is expected to increase significantly, this tool may help in decreasing the complexity requirements of decoders. In this work, the potential savings are analyzed for intra only coding and reach 5.74% of energy savings at the expense of a bitrate increase of 7.57% or 1.1% energy savings at 0.13% bitrate increase. The authors claim that due to the high amount of new coding tools proposed for VVC, energy savings of a higher magnitude at lower rate increases can be expected for inter prediction coding tools. It was asked whether the energy parameters that were used were provided. The presenter said they would provide these in a revision of the contribution. Further study was encouraged. It was noted that the “green MPEG” initiative seems related.
This contribution was provided for information on how to use memory bandwidth analysis tool. It is available for study. Detailed presentation was not requested. JVET-K0452 AHG5: Proposal of template for comparing memory bandwidth [R. . Hashimoto, S. . Mochizuki] [late] This was discussed Sunday 15 July 1630 (chaired by GJS & JRO). This contribution provides a template to compare memory bandwidth measured with the VTM and BMS software. This template can be used to check and compare memory bandwidth in the test model decoders. See the prior document JVET-J0090. The software is being integrated into the VTM and BMS software packages. It was commented that it could be desirable to request the use of the tool in AHG13 for tool on/off testing.
Yüklə 4,04 Mb. Dostları ilə paylaş:
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||