International organisation for standardisation organisation internationale de normalisation
Yüklə 8,63 Mb.
|
- Bu səhifədəki naviqasiya:
- TM Implementation
- CDVS Development and Requirements
- WD and TM Development
- Internet Video Coding
Test Model Development4.2.0.1.1.1.1.1.58m28125 CDVS: Test Model code maintenance [Massimo Balestri, Gianluca Francini, Skjalg Lepsoy, Andrea Varesio] (Chaired and recorded by Danilo Pau) The contribution examined TM4, in terms of portability and performance criteria. There was reported to be a problem in annotation in 1b and 1c execution speed extraction and retrieval, which were not finished. The performance achieved thus far was presented:
Conclusion: Create maintenance branch w/out affecting main trunk and adopt into TM5.0 without changing results. Switch to compute approximated and actual Jaccard index and correct annotation for 1a,b,c. It was proposed to insert these changes at first during TM5 integration. The group agreed to incorporate the proposed changes. 4.2.0.1.1.1.1.1.59m28296 Stripe Based DoG Hardware: performance evaluation in the TM4.0 [Mario Vigillar, Gian Domenico Licciardo, Danilo Pau, Arcangelo Bruna, Claudio Parella] This contribution proposed a stripe based DoG – avoiding full frame based buffering – total buffer size 512 kBytes – 14 bpp in 28 nm design (low leakage), 90fps @ 333MHz. Good performance – very slight (insignificant) drop in performance. TM Software Integration Plan
TM and anchors were planned to be released on 15 February 2013.
4.2.0.1.1.1.1.1.60m28232 Additional requirements on CDVS in Video Applications [Sabino Metta, Alberto Messina] This proposal was for possible requirements derived from extension of CVDS use cases to video content. All such requirements (combined) should be submitted to the Requirements group for consideraion.
New documents: WD 3 (N13334) and TM5 (N13335) Changes:
Code optimization (only software change) Update to accuracy measure of localization (slightly worse results, but believed to be more reliable) Software and anchor results were to be available Feb. 15
The following set of CEs was decided to be performed by the next meeting (see N13336):
A breakout group on IVC was active during the week to perform the actions regarding the following:
Regarding the alternative method, which had not originally been proposed in the CfP, it was suggested that a more detailed analysis should be performed for one more meeting cycle, with a decision for actions to be made at the next meeting. Besides the performance, the general status in terms of licensing, conditions of transfer to MPEG, software, and copyright require clarification and better understanding. To achieve the appropriate results in the comparison of ITM4 versus VP8 in an open process during the subsequent AHG period, CE1 (N13354) was established which includes plans for the codec settings, timelines, etc. Another CE, identified as CE2 (N13355) was set up which is intended to investigate technology for potential further improvements of ITM4 (not to be used in CE1). As part of the breakout activity, an expert viewing was performed on Thursday afternoon. By viewing VP8 and AVC anchor (from the CfP) results, it was confirmed that the data rate points planned for CE1 would be suitable for the formal subjective tests that are planned to be performed prior to the last meeting. Due to the lack of decoded sequences of either ITM3 or with added tools newly integrated in ITM4, no preliminary comparison was possible involving the ITM codec. Some further information regarding the licensing situation was provided during the video closing plenary (see under section 12) 4.2.0.1.1.1.1.1.61m27964 Adaptive Non-Reference P Optimization for Internet Video Coding [Xianguo Zhang, Long Zhao, Hao Lv, Ronggang Wang, Siwei Ma, Tiejun Huang, Wen Gao] From BoG:
In the Video plenary, it was asked whether subjective quality had been checked. With too drastic change of QP over the levels, there could be subjective degradation. 4.2.0.1.1.1.1.1.62m28161 cross-check of m27964 [Weiran Li] 4.2.0.1.1.1.1.1.63m27965 A direction and distance based motion vector prediction method for internet video coding [Hao Lv, Ronggang Wang, Lei Chen, Xianguo Zhang, Siwei Ma, Tiejun Huang, Wen Gao] From BoG:
It was suggested that delegates who bring questions regarding IPR on a proposal should provide the basis for that question (as in some evidence that leads them to question the IPR statement made by the proponents). It was suggested to draft a resolution requesting review of the IVC ITM for IPR ownership of the technology used. The BoG recommended to adopt this into the ITM after proposed text has been reviewed.
It wsa confirmed that the prediction is used for coding, but there is no signalling of candidates (rather, rules are established about selecting two candidates for computing an average). One suggestion was given that the operation suggested to compute the prediction (based on comparing candidates against each other) might be similar to median computation, the latter being potentially critical w.r.t. IPR. 4.2.0.1.1.1.1.1.64m28162 cross-check of m27965 [Weiran Li] 4.2.0.1.1.1.1.1.65m27966 New B frame encoding technology for internet video coding [Lei Chen, Ronggang Wang, Hao Lv, Xianguo Zhang, Siwei Ma, Tiejun Huang, Wen Gao] From BoG: This was a proposal to remove B_direct_16x16 and replace B_direct_8x8 with B_skip_8x8. It has a small gain of 0.3%. Discussion: this is done because other modes seem to cover the test sequences sufficiently. Reducing the modes reduces complexity of the codec. The BoG recommended to adopt into the ITM after proposed text has been reviewed.
From BoG: Different DCT formulation, based on Chen IDCT. Theer was an objection to adopting this proposal – requesting to require frequency response of the proposed transform before adopting (dropped). Proponents report that the motivation for this is that the ITM 3.0 transform has patent issues, so they went back to ITM 2.0 transform. There was an agreement to have informal viewing, but it was suggested that we probably have to adopt because of the patent issues in any case. Experts reported that the ITM 2.0 transform produced visual artefacts and that needs to be checked again. The BoG recommended to adopt this conditioned on the informal viewing results (no visual artefacts).
Discussed in conjunction with m28182, see the notes in section below. 4.2.0.1.1.1.1.1.70m28182 Comparison of the VP8 video codec and the current IVC Test Model [Lazar Bivolarsky, Mohammed Raad, Harald Alvestrand] Notes from the BoG (chaired and recorded by R. Wang): a. Comparison of VP8 to current IVC test model. b. Using VP8 encoder version 0.9.6 to ITM 3.0 c. Discusses the reportedly wide use and adoption of VP8 in the industry. d. Reports that VP8 (IPPP) has approx. 24% improvement over IVC ITM 3.0 (IPPP). e. Reports that VP8 (IPPP) has approx. 6% improvement over IVC ITM when used with IBBP. f. Recommends replacing the current ITM with VP8 (as described in m28187). i. Conditioned on the above, the authors recommended moving IVC to WD stage. ii. Prof. Yun He raised a concern regarding the patent situation of VP8. 1. The contribution states that the proponent is willing to license the proposed technology under Type 1 licensing. iii. There was interest from the group to look at the test scripts and visual results. Discussion was chaired by Ronggang Wang. Notes from Ronggang follow: Questions from the experts and Answers from the proponents: Q: Is there cross-checking for the performance of VP8? A: No cross-checking now. Q: Is there any text guidance for the VP8 encoder? A: No
Q: Where's the performance gain from? Please try to identify the key technologies. A: Entropy coding, segment organization, ... Q: If adopted, can it be changed in the future? Q: If it can't be changed, VP8 should go to ISO to be stamped as standard, other than contributing to IVC. Q: Are there any other third part techniques in VP8? A: Not sure. Q: If adopted and some modules are changed in the future, Google will still license for free for their IPs? A: Not sure, need to ask to Google lawyers Q: IPR analyses are required. Q: Verification of the performance gains needs to be done before the decision of changing platform. Q: What’s the motion vector prediction method in VP8? Q: Detailed experiment results? Q: Is there rate control in the testing? A: Yes, VBR rate-control is used. Comments from experts: (1) The proponents shall provide cross-checking input document from a third party. (2) A VP8 encoder description document is required to help the participants of testing to understand the source code of VP8. (3) Rate control shall not be used for performance testing. (4) Technique details on VP8 codec should be presented from proponents, not just providing a specification of VP8. (5) The IPRs in VP8 not owned by Google should be carefully checked. In further review, the points were not agreed upon by other experts. 4.2.0.1.1.1.1.1.71m28184 New IVC Test Sequences and AVC anchors [Lazar Bivolarsky] Notes from BoG: New sequences were proposed. i. We will need a common test set document. ii. Bitstream generation was to be checked during the week. iii. It was recommended to have this done by the end of the week. It was recommended to wait for IVC experts to check the new sequences during the week. It was recommended to adopt the new sequences if no issues were raised during this week. In later review, it was later agreed – after subjective viewing – that the new sequences as well as the rate points of the previous 720p / videoconferencing-type sequences are suitable. Therefore, they are included in CE1. 4.2.0.1.1.1.1.1.72m28107 Update on MPEG-IETF Coordination on Royalty Free Coding [Gerard Fernando, Rob Glidden] Not presented (as it was not obvious that it should be handled by the BoG). Other BoG activity (Monday and Tuesday): The BoG suggested a set of test conditions to use when testing VP8 vs. the current test model. The BoG was in agreement to plan for testing two different models and that this test will probably take place between now and the next meeting. Proponents for either technology had been asked to provide the configurations for each codec. Since we have the source code available to all members, all members have been invited to generate bitstreams for both codecs (to ensure that there is consistency). An informal expert viewing test was being organized with the help of the test chair (Vittorio) so that we could collect more performance-related information at this meeting, but it was understood that we may not get the time to fully complete this. A formal subjective test performed by Vittorio between the meetings was understood to potentially be feasible. The informal viewing took place in room T072 from 14:00 on Thursday the 24th of January. It was agreed that the new class D sequences provided in m28184 should be used instead of the class D sequences used in the IVC CfP. There was discussion in IVC breakout Thursday morning on further progressing (chaired and recorded by Jens-Rainer Ohm), as follows: Establishment of CE1: Investigation of overall compression performance Description shall include:
(It was later confirmed together with the Test SG chair that it is possible to run formal subjective tests prior to the next meeting. This is reflected in the CE1 plan and in the AHG mandates.) Some other discussion was performed on the usage of alternative reference pictures in VP8. It is confirmed by the contributors of 28187 that disabling multi-pass coding also disables usage of that feature. One other independent expert who has studied m28187 also confirms that this is the case. There is also the feature of “golden frame” which is a long-term reference picture. It was remarked by some experts that this may violate the random access constraint. It is however confirmed by the contributors of m28187 that this “GF” is reset whenever a RA position (keyframe) appears. The contributors of m28187 also announced that a more precise description of the encoder will be provided before the next meeting. CE2 (improvements of ITM) was reviewed. It will contain sub-experiments on:
|