Organisation internationale de normalisation
Yüklə 5,54 Mb.
|
- Bu səhifədəki naviqasiya:
- IVC Performance
- IVC development
- Output docs
- Reconfigurable Media Coding – Video related
- HDR/WCG Video Coding
This was suggested to be an encoder issue, trying to optimize the quantized DC coefficient of intra chroma blocks with SAD minimization. BoG Recommendation: To adopt it in the next version of ITM and encourage the contributors to do further investigations on full RDOQ on chroma components. 1.1.1.1.1.1.1.1.42m36763 Cross-check of m36710 on chroma enhancement technique on the intra predicted block for IVC encoding [Jae-Gon Kim, Anna Yang] No new performance gain was reported from the input contributions, indicating that IVC has reached its climax.
1.1.1.1.1.1.1.1.43m36681 Performance evaluation of Internet Video Coding [Xufeng Li, Ronggang Wang, Zhenyu Wang, Wenmin Wang, Siwei Ma, Tiejun Huang, Wen Gao] This document provides a comparison of AVC HP, WebVC, VCB and IVC Codecs based on BD bit rate under test conditions defined in N13943. In the RA constraint cases (CS1), IVC (as represented by the ITM software encoder) reportedly clearly outperforms the WVC (AVC CBP) encoder (as represented by the JM reference software encoder) and VCB in terms of BD bit rate numbers in overall average by 22.5% and 24% respectively, and reportedly underperforms AVC HP by 11.6%. In the LD constraint cases (CS2), IVC reportedly outperforms the WVC encoder and VCB by 15.8% and 6.5% in terms of BD bit rate respectively, and IVC underperforms AVC HP by 6.5%.
1.1.1.1.1.1.1.1.44m36781 Cross-check of m36681 (Performance evaluation of internet video coding) [Sang-hyo Park, Euee S. Jang] Tests were conducted Thursday morning, and the results of subjective viewing were reviewed by video Thu 06-25 15:30.
For output documents that were approvedby the video plenary, see further notes on the Thursday and Friday closing plenary discussions.
The following (continuing) Exploration Experiments are defined in N15365, which was edited by the BoG and approved in the video plenary. (Note: After approval of the CD, no new normative elements are planned to be added to the draft standard. The primary purpose of experiments is for improving encoder decisions, and investigate potential new elements for future extended versions of IVC).
See further notes about IVC (regarding actions after assessing the current quality in comparison against AVC High Profile) in Sec. 13. The following output documents were created:
Documents were presented during the Wednesday Video SG plenary. The following aspects were discussed in this context:
1.1.1.1.1.1.1.1.45m36653 Towards a parallel scalable profile of the dataflow HEVC decoder [Khaled Jerbi, Wassim Hamidouche, Mattavelli Marco] In the implementation, the FU for upsampling filtering is included in the enhancement layer decoder. Generally (in cases of spatial and SNR scalability), the EL decoder needs to be synchronized with the BL decoder at the CTU level, such that its operation could be slowed down, parallelism is eventually affected, or an additional picture buffer would be necessary between base and enhancement layer (note that HEVC does not define the upsampled base layer picture as being contained in the DPB of the enhancement layer). FUs still missing: MV upscaling, synchronization. PDAM expected for next meeting.
This is a report on the status of conformance test development. The software implementation is not fully conformant with main profile, as it does implement IPCM, and conformance streams related to special display modes (cropping) are not passed. One NB has commented about that. 1.1.1.1.1.1.1.1.47m36498 Update of the White Paper on Reconfigurable Media coding [Marco Mattavelli, Eduardo Juarez, Euvgeny Upnik, Alejo Arias, Simone Casale Brunet, Endri Bezati]
The Call for Evidence on high-dynamic range video had been issued by the Requirements SG, with responses due at the 112th meeting. Nine full submissions were received, which were initially investigated in the AHG and Requirements. During the week, the work was transferred to the Video SG, to coordinate further investigation. The mandate is for developing a solution that would enable usage of decoders compliant with Main 10 profile of HEVC (with no changes to the coding layer), with appropriate postprocessing of the decoder output (and complementary preprocessing at encoder input) such that it could be rendered to an HDR display. Backward compatible solutions (with the decoder output directly displayable on an SDR display with appropriate quality) shall also be investigated. It is yet to be determined which additional elements would be required in the specification (or if the current elements, e.g. for tone mapping, are sufficient), and whether specific profile definition would be necessary. In this context, it shall also be investigated whether the HEVC anchors that had been used in the CfE can be further improved, e.g. by further encoder optimization or optimization of tone mapping functions. To make progress towards these goals, three Core Experiments were set up (N15455/N15456/N15457), development of a software framework was started, and a first draft document describing a possible architecture of an HDR codec and processing chain was drafted (N15454). The following documents were initially presented in the AHG and the Requirements SG. 1.1.1.1.1.1.1.1.48m36811 AHG on HDR and WCG [Convenor]
This contribution reports the details and results of the subjective evaluations conducted at EPFL to evaluate the responses to the Call for Evidence (CfE) for HDR and WCG Video Coding. In total, nine submissions, five for Category 1 and four for Category 3a, were compared to the HEVC Main 10 Profile based Anchor. The subjective evaluations were conducted following the methodology presented during the AhG meeting in Lausanne in May. Five HDR video contents, compressed at four bit rates by each proponent responding to the CfE, were used in the subjective evaluations. Subjective results show that, on some contents, there is evidence that some proponents achieve statistically significantly better visual quality than the Anchor. 1.1.1.1.1.1.1.1.52m36835 HDR CfE Subjective Evaluations at Rome [Vittorio Baroncini] The formal subjective evaluation of HDR video clips requires a careful selection of the test method in relation to the kind of impairment that may be expected when compressing HDR video material. A decision was taken to use DSIS Side by Side comparisons with two repetitions. 1.1.1.1.1.1.1.1.53m36836 Results of HDR CfE subjective evaluations conducted at Rome [Vittorio Baroncini] This document describes the procedures and the results of the subjective evaluations conducted at Rome to subjectively assess the responses to the Call for Evidence (CfE) for HDR and WCG Video Coding. Nine submissions, five for Category 1 and four for Category 3a, were received and evaluated together with an Anchor obtained encoding the selected video content by means of HEVC Main 10 Profile. The formal subjective evaluation tests were done using a side by side version of the DSIS method. The test in Rome assessed four HDR video content (SRC), encoded at four bit rates. The result of the test showed evidence of superiority some submission that showed better visual quality than the Anchor at the same bit rate. 1.1.1.1.1.1.1.1.54m36184 Ericsson's response to CfE for HDR and WCG [Jacob Strom, Jonatan Samuelsson, Martin Pettersson, Kenneth Andersson, Per Wennersten, Rickard Sjöberg] This document describes Ericsson’s response to the CfE for HDR and WCG. It is a category 3a proposal and is therefore compatible with the HEVC Main10 profile. The processing follows that of the anchor generation process, with four changes. First, the calculation of the Y’ component is modified in order to avoid the luminance errors that occur in the anchor chain. Second, a different way to calculate the Cb and Cr components is used. Third, an anti-banding filter is employed in the post-processing before display. Fourth, the way to calculate the lambdas from the QPs is changed in the HM compared to how lambdas are calculated from the QPs in the anchor chain. The average BD rates for tPSNRY are reported as -3.4% and -9.3% for the BT.2020 and BT.709 test sets, respectively. For deltaE the average BD rates are reportedly -7.7% and -7.8% for the BT.2020 and BT.709 test sets. 1.1.1.1.1.1.1.1.55m36249 BBC’s response to CfE for HDR Video Coding (Category 3a) [A. Cotton, T. Borer, M. Pindoria, S. Thompson, M. Naccari, S. Schwarz, M. Mrak] This contribution describes a “Hybrid Log-Gamma” approach for HDR Production and Distribution. The approach has reportedly been mainly developed to address the needs of “live” TV production, as well as TV playout, presentation and distribution. It reportedly requires no end-to-end mastering metadata, which has reportedly proved difficult to support and unreliable in television environments. Furthermore, as it is based on an extension to the existing ITU-R Recommendations BT.709 and BT.2020, it is reportedly compatible with existing infrastructure and distributions systems and reportedly delivers a high-quality compatible image to legacy SDR displays. The quality of the compatible SDR image may not meet the requirements of the most critical applications, where only transmission of separate manually graded HDR and SDR assets will suffice, but is reportedly sufficient to address the needs of television broadcast where distributing two independent versions of a programme, or even using scalable coding solutions, have reportedly proved impractical in the past. 1.1.1.1.1.1.1.1.56m36251 An Efficient Dual Stream Approach to HDR Video Coding (Cat. 1) [W. Dai, M. Krishnan, P. Topiwala (FastVDO)] FastVDO studied splitting the High bit-depth (HBD) or floating point video signal processing into two coded streams: a smoothed luminance (SL) signal, and an associated base signal (B). High Dynamic Range (HDR) and Wide Colour Gamut (WCG) video content is first converted into an integer colour space called YFbFr, designed to be closely aligned to the YCbCr colour space of BT.709, that separates luma and chroma components. From the Y component, a grayscale smoothed luma signal (SL) is generated, and then used to generate a low bit-depth Base signal (B). Both signals are then converted to YFbFr 4:2:0 signals, and coded using the Main10 High Efficiency Video Coding (HEVC) profile. 1.1.1.1.1.1.1.1.57m36256 Single layer non-normative (category 3a) NCL and CL responses to the Call for Evidence on HDR/WCG [D. Rusanovskyy, S. Lee, D. Bugdayci, A. Ramasubramonian, J. Sole, M. Karczewicz, A.M. Tourapis, Y. Su, D. Singer, C. Fogg, A. Duenas, F. Bossen] This document presents two responses to the MPEG Call for Evidence on HDR/WCG coding technologies. Both responses use the HEVC Main 10 profile with identical encoding configurations. The first uses a conventional framework for video coding, the non-constant luminance (NCL) representation, and the second employs a constant luminance (CL) approach. The tools presented in these responses are reportedly non-normative to the video coding process. Gains in terms the objective metrics used for the analysis are reported for both responses against the CfE anchor. Some visual improvement over the CfE anchors is also reported.
It is suggested to classify HDR video compression into two approaches: one-stream approaches that take advantage of the higher bit-depth support in the latest video encoders, and two-stream methods which can exploit legacy encoders and newer higher bit-depth techniques. Falling within Category 1 of the MPEG Call for Evidence, one-stream methods transform a single HDR input stream into a single compressed stream using some form of transfer function. This transfer function can affect the quality of the resultant decompressed steam, and its decoding complexity. A one-stream method is described which can operate at high frames, e.g. 120fps. 1.1.1.1.1.1.1.1.59m36262 Optimex: A Category 2b HDR video compression method [Kurt Debattista, Jonathan Hatchett, Elmedin Selmanovic, Tom Bashford Rogers, Ratnajit Mukherjee, Joshua McNamee, Alan Chalmers] Two-stream HDR video compression methods split the input HDR video stream into some form of base and detail layer, which are each then compressed. A backward compatible two-stream method, which falls within Category 2b of the MPEG Call for Evidence, reportedly ensures that the base layer may be viewed with legacy SDR players. A tone mapping operator (TMO) is used to display the HDR stream on SDR displays. Although various TMOs have been proposed over the years, which is best for any content is reportedly subjective. This document presents a backward compatible two-stream HDR video compression method which identifies a single exposure that is used to display the SDR stream. It is reported that such an approach can be adapted to fit within the bit depth of any traditional encoder, is able to reproduce the most information from the original HDR image, is straightforward to implement and fast, and avoids the problem of deciding on an appropriate TMO and its associated parameters. 1.1.1.1.1.1.1.1.60m36263 Technicolor's response to CfE for HDR and WCG (category 1) [S. Lasserre, F. Le Léannec, E. François, T. Poirier] This contribution describes the Technicolor response to the MPEG Call for Evidence for HDR and WCG Video Coding for category 1 related to normative changes to the HEVC specification. The approach is based on a single layer codec design and can provide SDR backward compatibility, by using a pre-processing step applied prior to the encoding. The resulting SDR video can be compressed then decoded using legacy decoders (e.g HEVC Main 10 decoders) and directly rendered on SDR displays. Dynamic metadata of limited size (coded per picture, GOP or scene) are used to reconstruct the HDR signal from the decoded SDR video, using a post-processing that is the functional inverse of the pre-processing. Both HDR quality and artistic intent are reportedly preserved. Pre- and post-processing are applied independently per picture, do not involve any inter-sample dependency, and are codec independent. 1.1.1.1.1.1.1.1.61m36264 Response to Call for Evidence for HDR and WCG Video Coding: Arris, Dolby and InterDigital [D. Baylon, Z. Gu, A. Luthra, K. Minoo, P. Yin, F. Pu, T. Lu, T. Chen, W. Husak, Y. He, L. Kerofsky, Y. Ye, B. Yi] This proposal is in response to the Call for Evidence (CfE) for HDR and WCG Video Coding, resulting from collaboration among Arris, Dolby and InterDigital. The response falls in Category 1: compression efficiency improvement over HEVC Main 10 Profile for HDR with BT.709 or P3D65 Colour Gamut Content. It is asserted that the coding technologies in this proposal provide sufficient evidence that the coding efficiency can be significantly improved over the High Efficiency Video Coding (HEVC) Main 10 Anchors for HDR and WCG video content. The proponents recommended that a Call for Proposals be issued as soon as feasible. 1.1.1.1.1.1.1.1.62m36266 Philips response to CfE for HDR and WCG [Rocco Goris, Robert Brondijk, René van der Vleuten] This document describes the Philips response to the Call for Evidence (CfE) for HDR and WCG. Philips submitted a response in the “compression efficiency improvement” category. The proposed compression method provides a form of backward-compatible SDR video. 1.1.1.1.1.1.1.1.63m36280 Dual layer non-normative (category 3b) response to the Call for Evidence on HDR/WCG [A.K. Ramasubramonian, J. Sole, M. Karczewicz, S. Lee, D. Rusanovskyy, D. Bugdayci, X. Li] During the week, breakout meetings were held every day until Friday morning. Meetings with presence of a video co-chair (JRO) were Wed 14:00-18:00 and Thu 11:00-13:00. During these meetings, the following conclusions were drawn. The subsequent action items (setup of Core Experiments, document of common technology and architecture concepts found from the CfE results) were approved by the video subgroup in the Friday plenary. Work action items and high-level conclusions:
Work plan
Output doc and interim scheduling plans
Notes from the video and BoG meetings about document review: Early agreements were that there could be two anchors in subsequent investigations. One that is the typical MPEG anchors using common test conditions (as in the CfE). Another further optimized anchor termed the “super anchor” could act as a proponent. 1.1.1.1.1.1.1.1.64m36271 On 4:4:4 to 4:2:0 conversion - Performance of downsampling and upsampling filters, and MinMax and Closed Loop Filtering [A.M. Tourapis, Y. Su, D. Singer (Apple), J. Sole, D. Rusanovskyy, S. Lee, D. Rusanovskyy, A. Ramasubramonian, M. Karczewicz (Qualcomm), C. Fogg (Movielabs), A. Duenas, F. Bossen (NGCodec)] This contribution tries to address some issues on chroma format conversion as they were observed within the MPEG HDR/WCG activity. It is suggested that these issues do not only impact HDR/WCG material but also the conversion of SDR material. Considerable benefits, both objective and subjective, could reportedly be achieved using improved conversion techniques, as well as techniques that account for the characteristics and the representation of the material. The suggestion is to replace the 1,6,1 filters used in the CfE by various other filters. It is attempted to avoid filters with negative taps. Up to 1.60 dB improvement in tPSNR was reported. Video viewing done by Alexis Tourapis reportedly showed improvement in visual quality. Alexis Tourapis, Jacob Strom and Jonathan Samuelson were asked to further study, create a new software package in HDRtools and release that before the next meeting. 1.1.1.1.1.1.1.1.65m36272 Clipping during 4:2:0 to 4:4:4 conversions [A.M. Tourapis, Y. Su, D. Singer (Apple), J. Sole, D. Rusanovskyy, M. Karczewicz (Qualcomm), C. Fogg (Movielabs), A. Duenas, F. Bossen (NGCodec)] This contribution provides informative recommendations relating to chroma processing, especially pertaining to the Non-Constant Luminance (NCL) and Constant-Luminance (CL) representations. In particular, it is suggested that when processing chroma information, e.g. for the purpose of chroma format conversion or rescaling, among others, any necessary clipping should be performed while also taking in account the relationships of and information from all available colour planes. This can reportedly help in reducing conversion artefacts and in improving subjective quality. Comments: This has smaller possible gain in picture quality but may avoid some of issues associated with conversion and clipping artefacts that may appear in some content. It can be applied to any colour space. Code is implemented only for NCL, so far. Some default setting of the software is needed (as the number of flags is increasing to a large number. Action: This was added in the HDRtools software and made available but not fully released yet. It will be made part of the software release associated with M36271.
One of the primary artefacts that have been observed on the Anchor material generated for the MPEG HDR/WCG CfE, was the presence of blockiness, especially on material with considerable motion and noise. Such blockiness may also impair the performance of different chroma conversion techniques, i.e. for up-converting from a 4:2:0 to a 4:4:4 representation, given the additional arteficial high-frequencies introduced into the content. This contribution suggests that adding an additional chroma-only deblocking filter, after decoding and before upconversion to 4:4:4, could considerably improve both the subjective and objective performance of HDR, and material using Main 10 profile encoding. Action: Implemented in private branch of HDRtools. It will be made part of the software release associated with m36271 and m36272 1.1.1.1.1.1.1.1.67m36376 SHVC encoder improvement [P. Bordes, K. Y. Cai, F. Hiron, E. François] At the Geneva February 2015 meeting (the 111th MPEG meeting), it was reportedly observed that some reconstructed pictures of the SHVC anchors of HDR-WCG CfE may have some kind of colour artefacts. Modifications to the SHM encoder were proposed here, reportedly allowing removing most of these artefacts. The enhancement layer (EL) bit-rate is increased (5% to 26% in average depending of the test sequences) but the PSNR and the subjective quality is reportedly unchanged. A colour issue occurs due to transitions in between octants. A fix is proposed that detects the problem and selects a different set of values. There is a slight objective performance decrease. Further study is encouraged since it is not time critical. Encoder optimization for the SHVC anchors was described. The contribution reports that artefacts could occur when SHVC CGS is used for HDR scalability, and suggests “artefact detection” based on colour saturation. In that case, CGS is disabled. Bit rate inreases by approximately 5% on average were reported – more on some sequences. Artefacts are mostly visible at high bit rates. It was asked whether this has an impact on bit rate fluctuation, which should be further studied. There were no CfE contributions in cat. 2a (scalable approaches). Further study was recommended to further improve SHVC, but no urgent action was required. 1.1.1.1.1.1.1.1.68m36434 Upsampling and downsampling filters for HDR content [A. K. Ramasubramonian, D. Rusanovskyy, J. Sole, M. Karczewicz, S. Lee, D. Bugdayci Sansli, J. Chen] This proposal documents several chroma upsampling and downsampling filters that are reported to provide objective gains in coding HDR content. Different pairs of upsampling and downsampling filters are reported to be used with different sequences. Objective gains of 1.9% in deltaE BD-rate were reported for the HDR test sequences. It is proposed to allow signalling of the upsampling and downsampling filters in SEI messages to enable decoders to choose a filter to decode the bitstreams that would provide good objective performance. Test of several alternative downsampling and upsampling filters (Lanczos, DCT based, for latter case re-using existing HEVC filters) It is said that the existing chroma resampling SEI message is not suitable for the given case, because its persistence scope is the entire CVS. However, in the contribution, no advantage was shown for changing the resampling on a per-picture basis – the filters were selected just once per sequence. The proposal is specific for HDR, but it was suggested that it might be more relevant in the HDR case than for SDR. It is pointed out that the post filter hint SEI message could also be invoked to further improve the quality of the output. Further study was necessary, whether the chroma resampling SEI message could be used for further improving anchors, or whether modifications are needed for more frequent dynamic changes. This could fit into the CE1 context, but there are higher priority aspects. No direct action was taken on this. Further study was encouraged.
This document shows coding results of a Y''u''v'' (in short Yuv'') system, as first presented in document M35065, for the content defined in the CfE. In the system used to process the CfE content, the Philips EOTF at 10000 NIT was used in order to reach the same low QP values as the anchors do. This reportedly assures that the compression artefacts in Yuv” are about same as that of the anchors. This “Double prime” approach as as follows: u’v’ components are generated from a linear space, and after downsampling, further nonlinearly is adapted into u’’v’’ (mainly effective in the “high nits” range). Y is transformed with an EOTF. Some de-noising of chrominance is also in effect in this context, since clipping is performed for low Y values. After decoding 4:2:0, the process is reversed. An asserted advantage is that encoding is performed in a constant luminance domain. It was pointed out by other experts that some of the objective results may indicate that the material is more difficult to encode. The proponent reports however that subjective improvements were observed at the same bit rates. The proposal uses a different mapping not currently supported in HEVC; this would fit in the CE2 context. The signal still contains some Y "cross contamination". The question was raised if there should be a CL version of the anchors. Chad Fogg and Ericsson will create CL anchors to compare and study further. Additional CL anchors should be generated in CE1 (not from this proposal, but to allow comparison with other CL methods). 1.1.1.1.1.1.1.1.70m36462 Additional Results for m36184 for Colour Correction [Jacob Ström, Martin Pettersson, Jonatan Samuelsson, Kenneth Andersson] This contribution shows that saturation artefacts appear in the anchors when the BT.709 "container" is used, even in the uncompressed case. This is however not a problem of the container, but rather the way of processing that is used to generate the data for the container. The problem is a combination of the use of non-constant luminance, 4:2:0 subsampling and the non-linear transfer characteristics function. It should be further investigated in CE1 how this problem can be solved (other contributions pointed to similar problems). 1.1.1.1.1.1.1.1.71m36463 Reporting: Coding efficiency of the scalable coding for HDR and SDR video using inverse tone mapping on SHVC [Jeongyun Lim, Woong Lim, Yong-Jo Ahn, Seanea Park, Donggyu Sim] This approach applies an inverse tone mapping function (a global function) in the base-to-enhancement prediction of SHVC, which is used instead of CGS. Compared to simulcast, it reportedly achieves approximately 15% total bit rate reduction (SDR plus HDR), and approximately 45% in the EL (HDR) only. The base and enhancement layers used the same resolution. No comparison was performed against the SHVC/CGS anchors. No immediate action was taken on this. 1.1.1.1.1.1.1.1.72m36274 Coding results of noise-reduced 4:2:0 and 4:4:4 HDR signals [K. Slavin, C. Fogg] Video clips being studied in the HDR Call for Evidence were noise reduced prior to HEVC encoding in order to provide a visual comparison with the same clips coded without a noise reduction pre-processing step. No changes were applied to the HEVC encoding or decoding steps in this test. Detail and sharpness were maintained by using a 16-bit per sample motion-compensated filter. Playback of the bitstreams was demonstrated visually at the meeting. The algorithm uses integers. Chad Fogg looked at images himself since the algorithm has no identification of noise verses structure. Gains were reported on some sequences at the higher PSNR values (around 50 dB). Indications reportedly were that 4:2:0 and 4:4:4 showed improvements. No immediate action was taken on this.
The authors of this information document report that High Dynamic Range / Wide Colour Gamut video content can be or has been exchanged between single-layer HEVC Main10 products. A description of the metadata and example settings for open source software encoders was provided. This was an information document on HDR Main 10 usage. No Action was taken on this.
This contribution suggests that workflows need at least 28 bits of precision (RGB linear light) for HDR. No action was taken on this. 1.1.1.1.1.1.1.1.75m36278 Output code map SEI for fractional internal bit-depth increase [C. Fogg (MovieLabs), B. Mandel (Universal), A. Tourapis, Y. Su, D. Singer (Apple)] OCM proposal – was reviewed at the last meeting. 1.1.1.1.1.1.1.1.76m36279 Wiener filter for noise reduction pre-processing of HDR video signals [A. Tourapis, C. Fogg] A noise reduction Wiener filter was added to a beta version of the HDRTools C++ source software program being used in the HDR/WCG adhoc study. The filter currently operates directly upon .EXR files (with half-float samples), typically containing linear light domain video RGB primary video signals. Denoising is not in the primary focus of necessary actions currently, and no time was left during the week to discuss this again. No action was taken on this.
This contribution proposes a new SEI message for carriage of metadata for use in post processing stages of HEVC-based HDR/WCG content delivery. The name of the proposed SEI message is the "dynamic range adjustment SEI message". This reuses a remapping of sample values. There may be some issues with the proposed syntax. It was suggested to improve the anchor and so have this go to CE1. Anther suggestion was to consider putting it into CE2 if applicable.
This contribution proposes to support the Colour Remapping Information (CRI) SEI message, which has been included in the HEVC specification, in the AVC standard as well, in order to leverage CRI-based technologies for AVC-based services. The CRI SEI message was released with HEVC version 2. CRI applies when two reference master versions of a same content (e.g. SDR and HDR masters) are produced. The CRI SEI message was designed to enable content conversion from one version to another one (e.g. HDR-to-SDR), with respect to a master visual rendering. It was reported that HEVC CRI SEI message is included in the UltraHD Blu-ray specification for conveying dynamic metadata enabling an optional reference-based HDR-to-SDR down-conversion. Objective results of CRI-enabled HDR-to-SDR down-conversion computed on MPEG HDR test material were provided. The proponent asked to include the CRI SEI message into the AVC standard. No immediate action was taken on this suggestion.
In M34274, Philips proposed an EOTF for HDR video. It was suggested that since a framework for evaluation has now been defined for the CfE, we can evaluate the proposed EOTF by replacing the current SMPTE 2084 EOTF in the anchor generation chain by the other proposed EOTF. Specifically, the contributor proposed an EOTF that is reportedly optimized for ���� = 5 000 cd/m2 for which the constants are �� = 25 and �� = 2.4. The lower peak luminance is asserted to provide better coding efficiency due to decreasing the granularity of codeword usage. The same can done by remapping pixels.This will be studuied in CE1.
This document presents the effects of enabling chroma QP offset usage for Ericsson’s response to the CfE on HDR and WCG. It can be noted that Ericsson’s response to the CfE on HDR and WCG is compatible with the anchor processing chain and only modifies the calculation of Y’, Cb and Cr to reduce visual artefacts of the anchor, while making a modification on the lambda calculation in HM to perform improved RDO decisions and applying an (optional) anti-banding filter in case the display supports a bit-depth higher than 10 bits. These modifications reportedly do not have a direct impact on the relative quality between luma and chroma. With a negative chroma QP offset, more bits are spent on chroma relative to luma, which reportedly improves BD rate performance for deltaE. Informal viewings also reportedly indicate that the subjective quality is improved. It is asserted that the effect of applying chroma QP offset would be similar for the anchor chain as for Ericsson’s response. The contribution recommends using Chroma QP offsets. Luma begins to suffer as more Chroma QP offset is used. This will be studied in CE1.
This document is HDR/WCG CfE-related and provides additional information on HDR/WCG coding approaches presented in document m36256. This coding configuration was developed and tested in the CfE framework and is an extension of the CfE responses presented in m36256. Current results were not included in CfE response due hardware problems during the bit-rate matching process. The proposed coding technique reportedly provides further improvement in coding efficiency compare to reference methods in m36256. For the NCL-response, the proposed method reportedly brings about 43%, 19% and 10% of BD-rate improvement in deltaE, mPSNR and tPSNR_XYZ metric against the anchor, respectively. For the CL-response, the proposed method reportedly brings about 39%, 15% and 12% of BD-rate improvement in aforementioned metrics, respectively.
This contributions reports that Rec. ITU-R BT.2020 appears to prescribe a different chroma positioning for 4:2:0 sampling than what has been previously assumed in video coding work. Based on the chroma sampling location type numbering scheme found in AVC and HEVC, it is reported that, rather than using location type #0 (which is considered the default in AVC and HEVC and is prescribed in MPEG-2) or type #1 (which is considered the default in H.261, H.263, and JFIF), it appears that (in its Table 5) BT.2020 specifies the use of a 4:2:0 chroma sampling lattice with chroma sampling location type #2. BT.2020 uses a fixed location for chroma samples. This raises the issue that the SHVC spatial scaling design is conceptually incorrect due to its assumption of a different chroma positioning, which may result in some degradation of compression performance. After discussion and coordination, it was agreed that a liaison statement would be sent by ITU-T Q6/16 to ITU-R WP6 to inquire about this issue. 1.1.1.1.1.1.1.1.83m36749 Use of chroma QP offsets on HDR/WCG sequences [A.M.Tourapis, Y. Su, D. Singer, C. Fogg, A. Duenas, F. Bossen, J. Sole, D. Rusanovskyy, S. Lee, D. Bugdayci, A. Ramasubramonian, M. Karczewicz] Several responses to the MPEG HDR/WCG Call for Evidence (CfE) seem to employ a fixed or dynamic scaling of the colour components prior to compression. This document suggests that, especially at the lower bit rates that the CfE has been targeting, a similar behaviour could be achieved by adjusting the quantization parameters in the encoder. Chroma QP offsets, for example, could be used to adjust for a difference in the distribution of bits between luma and chroma transform coefficients. Objective gains on some of the CfE test sequences were shown. It was also pointed out that dynamic adjustment of all colours, which basically works as dynamic quantization in the sample domain, can also potentially be emulated using non-fixed quantization across all colour components. Adaptive quantization, at both the picture and CU levels, however is commonly employed in real encoding systems and is commonly out of scope within standardization efforts. This supports Ericsson data in another document. It will be studied in CE1.
Legacy television systems provide Standard Dynamic Range (SDR), supporting a range of brightness that is significantly smaller than the range that the human eye is capable of discerning. Similarly, current video systems do not support the wide range of colours that the human eye can perceive. Future television and other video distribution environments are expected to give a viewing experience that is closer to a real-life experience, to provide the user with a stronger sense of “being there”. This document describes suggested requirements and use cases for higher dynamic ranges and wider colour gamuts than are typically supported today. MPEG has initiated an effort to determine if any changes to the current MPEG standards are needed to meet these requirements. The document has been discussed. Changes in the proposed generic block diagram are suggested, to make the design more generic. E.g., the title of the box for conversion to 4:2:0 has be renamed to something more generic. There was discussion on the case of SDR backward compatibility and the evaluation of the subjective SDR quality. Regarding viewability in broadcasting applications, no reference is available. One discussed approach to evaluate SDR quality is to check if it respects the HDR rendering intent. Visual comparison with HDR displayed on an HDR monitor on one side, and SDR displayed on SDR monitor on another side was suggested, while asking viewers to assess the colour similarity between the two versions. Another option is to perform visual tests on the SDR version and compare MOS scores compared to scores of a graded SDR high quality reference. It was also suggested to try to define a MOS score threshold below which video quality is considered as not acceptable. Inputs are requested from those interested in this topic. Another discussed topic was where is conformance point. The authors of this contribution suggest to consider the HDR conformance point to be after final conversion to ST.2084 4:2:0. This was recommended by the authors to be used as starting point for further discussions for HDR requirements. It was suggested to re-submit the document at the next meeting for further requirements refinements. 1.1.1.1.1.1.1.1.85m36317 Modulation-based HDR video coding with SDR backward compatibility [S. Lasserre, F. Le Leannec, E. François, T. Poirier] This contribution presents a single-layer HDR video coding approach targeting SDR backward compatibility. The approach is based on a pre-processing step applied prior to the encoding that generates from the input HDR content an SDR version and some side modulation metadata of limited size. The SDR video can be coded/decoded using legacy codecs (e.g., compliant with HEVC Main 10) and the decoded video can be directly viewable on SDR displays. A new SEI message was proposed to transmit the modulation metadata (coded per picture, GOP or scene). Modulation metadata are used to reconstruct the HDR signal from the decoded SDR video, using a post-processing operation that is the functional inverse of the pre-processing. Both HDR quality and visual rendering are reported to be preserved. Pre- and post-processing are applied independently per picture, do not involve any inter-sample dependency, and are codec independent. This was submitted as a JCT-VC document (although it was not reviewed there).
This document was not uploaded and was considered as withdrawn. 1.1.1.1.1.1.1.1.87m36612 Upsampling and downsampling filters for HDR content [A.K. Ramasubramonian, D. Rusanovskyy, J. Sole, M. Karczewicz, S. Lee, D. Bugdayci Sansli, J. Chen] This proposal documents several chroma upsampling and downsampling filters that are reported to provide objective gains in coding HDR content. Different pairs of upsampling and downsampling filters are reported to be used with different video test sequences. Objective gains of 1.9% in deltaE BD-rate areis reported for the HDR test sequences. The contribution was for information only – no action was requested.
1.1.1.1.1.1.1.1.88m36243 Visualization of HDR metrics [L. Kerofsky, Y. Ye, Y. He] This document describes efforts to visualize the results of various objective quality metrics being considered for HDR work. During subjective testing, several subjects commented on seeing colour related artefacts when viewing the CfE anchor results. Colour blocking artefacts were noticed on neutral colour areas of some images. Colour artefacts such as banding were seen in the sky of one sequence and in a face image. The contribution was motivated by a desire to understand the ability of various objective metrics described in the CfE to detect such colour artefacts. A visual representation of each CfE metric was produced to examine the response to artefacts noted by subjective viewing. This visualization effort led to the assertion that both mPSNR and tPSNR were quite similar, i.e. equally bad, in being unable to identify the artefacts identified by the subjective viewers. The DE-2000 scheme was reportedly better in that some response to known artefacts could be seen. A new colour distortion measure inspired by the DE-2000 metric was proposed, called the Colour Delta E (CDE) metric. It was asserted that this CDE metric better measures to the colour distortion noted during subjective viewing.
This document describes a refinement of the metric first described in m35866, and gives evidence of its merit. 1.1.1.1.1.1.1.1.90m36600 On HDR Objective Metrics and Subjective Evaluation [Z. Gu, D. Baylon, K. Minoo, A. Luthra] This contribution presents some results comparing subjective evaluations of HDR test sequences with objective metrics applied to the same sequences with a focus on preference correctness. A group of viewers were shown side-by-side comparisons of sequences used in the HDR CfE and the sequences were ranked subjective quality using a 5 point scale. Objective metrics included the mPSNR, tPSNR, and deltaE_PSNR metrics used in the CfE as well as a HDR-VDP-2 metric. 1.1.1.1.1.1.1.1.91m36615 Preliminary study on HDR-VDP-2 for HDR/WCG [S. Daly, T. Lu, W. Husak (Dolby), L. Kerofsky] This contribution presents some preliminary study of the HDR-VDP-2 metric to explore its potential usage in helping develop HDR/WCG compression approaches. HDR-VDP-2 is in the class of calibrated quality metrics using physiological models. The preliminary study reportedly shows it can provide useful information in indicating and quantifying areas where texture loss can be perceived, which is helpful in the evaluation of HDR/WCG coding approaches. Further study on this metric was recommended. 1.1.1.1.1.1.1.1.92m36769 Distortion Metrics for HDR/WCG: Peak Overall Signal to Noise Ratio [A.M.Tourapis, Y. Su, D. Singer, C. Fogg] During the MPEG HDR/WCG Call for Evidence (CfE) activity several of the responders expressed the opinion that the majority, if not all, of the metrics used for objective evaluation failed in providing good correlation with, commonly their own, subjective evidence. A few participants, however, made the claim that the CIE DE2000 measurement may be a superior measurement relative to the others. This was reportedly primarily driven by the apparent and rather substantial Bjontegaard “gains” demonstrated for this measurement for almost all proposals, which though seemed not to be really justified by the subjective experiments. Without claiming to be solving the objective quality analysis problem, the contribution presents a modification to the tPSNR/PSNR computations and a regional distortion computation for tPSNR and possibly other measurements, that was implemented in the HDRTools package. Such measurements could be used for future analysis. 1.1.1.1.1.1.1.1.93m36270 HM reference software bug fixes and enhancements to address the HDR/WCG CfE [A.M. Tourapis, Y. Su, D. Singer, J. Sole, D. Rusanovskyy, S. Lee, D. Bugdayci, A. Ramasubramonian, M. Karczewicz, C. Fogg, A. Duenas, F. Bossen] This contribution presents several video encoder enhancements as well as bug fixes that were introduced into the HM reference software in work on the responses for the MPEG HDR/WCG CfE. The enhancements include a new, improved motion estimation technique, several new methods for weighted prediction, speedups to the distortion computation, and improvements to the rate distortion optimization, among others. Bugs that were impacting the performance and behaviour of the weighted prediction among others, were also reportedly discovered and fixed. The contribution suggests that these enhancements should be valuable not only for the MPEG HDR/WCG activity but also for future video codec development activities that might be undertaken for other purposes. It was suggested that these enhancements are also adopted into the official reference software. Some recommendations, in terms of common coding conditions, were also provided. This was reviewed in JCT-VC and the modifications were adopted into the HM reference software.
This contribution presents results of compression of two HDR HD sequences made available by the EBU to MPEG for the HDR-related exploratory activities. This will be considered for further review in a September AHG meeting.
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||