International organisation for standardisation organisation internationale de normalisation

4.6Closing session notes

In the closing session there were no requests to reopen discussions of preceding agenda topics and side activities recorded elsewhere in this report.
The JVT thanked its WG 11 parent body and Huawei for hosting the 25th JVT meeting.
The meeting was closed at 12:35 pm on Friday 26 October 2007.

4.7JVT liaison communications

The JVT did not receive liaison communications at this meeting.

The JVT sent liaison statements as found in JVT-Y201 to a number of organizations – specifically to BDA, DVD Forum, DVB, ATSC, SCTE, SMPTE, 3GPP, 3GPP2, OMA, ISMA, IETF. The purpose of these liaison statements was

• 
  to inform relevant organizations of the recently-completed work on professional profiles and scalable video coding, and
  
• 
  to solicit input from relevant organizations on requirements for future work on scalable video coding (SVC) relating to bit depth, chroma format, and color gamut issues.
  

5AVC base specification and related topics

The latest prior errata reporting status was provided in the JVT-X210 output document of the previous meeting. Two additional documents related to errata issues were submitted for consideration at this meeting: JVT-Y083, and JVT-Y084. One of these was considered by the JVT and the other was withdrawn by its contributor.

5.1.1.1.1JVT-Y083 ( Prop 2.2) [S. Narasimhan (Motorola)] On end_of_stream NAL unit and splicing

Splicing is currently used in US Cable networks for digital advertisement insertion based on the MPEG-2 video standard, and there are plans to migrate these applications based on AVC in the near future. In these applications, the splicing equipment (or function) combines two independently encoded AVC streams and is expected to produce an AVC “conforming” output for receiving equipment. This contribution outlined some issues related to generating an AVC conformant output by such splicing equipment and suggests some changes to the AVC standard to assist these applications.
JVT decision: In the next corrigendum, we plan to add an informative note to subclause 7.4.2.6 (end of stream RBSP semantics), stating approximately as follows:
“NOTE – When an end of stream NAL unit is present, the bitstream is considered to end (for purposes of the scope of this Recommendation | International Standard). In some system environments, another bitstream may follow after the bitstream that has ended, either immediately or at some time thereafter, possibly within the same communication channel.”
For no_output_of_prior_pics_flag (item 64 of JVT-X210), a “(but should not)” or “permitted (but not recommended)” is planned to go into subclause 7.4.3.3.
Our plan in relation to the ISO/IEC approval process, is to issue a DCOR in January and COR in July.
JVT has agreed on how to address end of stream NAL units and no_output_of_prior_pics_flag, and will document the details of the conclusion in an upcoming COR.

5.1.1.1.2JVT-Y084-L (Late Prop 2.2) [A. Rodriguez (Cisco)] Provisions for concatenation and splicing of streams <>

This contribution was withdrawn by its contributor.

6Scalable video coding (SVC) phase II

6.1Core experiment #1 & related docs: SVC bit depth scalability

6.1.1.1.1JVT-Y039 ( Prop 2.2/3.1) [M. Winken, H. Schwarz, T. Wiegand (HHI)] CE1: Results for SVC bit-depth scalability

This document presents results for the Viper sequences using configuration 1 and 3 of CE 1 on bit-depth scalability using the approach as described in JVT-X057. Results from the current design using latest test conditions were provided. No change to syntax, semantics, and decoding process. Encoder bug fixes and some other software progress were reported.
Includes PPS-level tone map specification – vs. prior SPS-level.
This method remains our current reference design.

6.1.1.1.2JVT-Y056-V ( Info) [M. Wien (Aachen U.)] Verif JVT-Y039 results for bit-depth scalability

This was a verification contribution relative to CE1 SVC on bit depth scalability for document JVT-Y039. The results of JVT-Y039 were reported to have been confirmed.
Summary of the verification task: The proponents reportedly provided the source code of the modified JSVM_8_12 software with BDS as distributed among the CE-participants, configuration files, simulation scripts, and bitstreams according to the test conditions.
For verification the following steps were reported to have been performed:

• 
  The software provided by the proponents was successfully compiled under Linux and was used for the simulations.
  
• 
  The test sequences provided on the ftp site at Sharp and provided by the proponent were used.
  
• 
  Simulations were run using the simulation scripts provided by the proponents.
  
• 
  The provided bitstreams were successfully decoded.
  
• 
  The results were reportedly found to exactly match the results reported in the RD data provided with the streams. These results reportedly matched the results reported in contribution JVT-Y039 for both configurations (configurations 1 and 3). The results were reported in the accompanying Excel sheet.
  

Question: How carefully was this software studied? Answer: Not much study – just ran the software.

6.1.1.1.3JVT-Y032 ( Prop 2.2/3.1) [M. Winken, H. Schwarz, T. Wiegand (HHI)] Bit-depth SVC: New results for Freeway sequence

This document presents new results for the Freeway sequence under the testing conditions as given in JVT-W302 (Core Experiment on Bit-depth Scalable Video Coding). It was reportedly shown how a significantly improved rate distortion performance could be achieved by using a three layered approach with an AVC-rewritable 8 bit intermediate layer. This approach reportedly requires no changes in syntax and semantics to the current JSVM. As a reference, the results for this sequence as summarized in JVT-X102 were given.
Shows improvement for this sequence (bad performance of proposal in results of last meeting). Claims that tone-mapping operator as used for this sequence produces low-contrast 8 bit sequence (higher bit levels hardly used). Now uses a three-layer approach (with linear 8-bit in between, using re-writing as one stream). Tweaked manually, but most probably an automatic algorithm could detect such weird tone mapping and automatically run such an approach.
Suggested interpretation of prior phenomenon of not-so-good performance on this sequence: interaction of two factors

• 
  non-linear tone-mapped picture rather dark (max luma 162), and
  
• 
  MCP was only using the 8-bit base layer.
  

Proposed solution: a 3-layer approach using an intermediate layer and final layer for the 10-bit enhancement with linear tone mapping. Two bottom layers structure to enable AVC rewriting.

Modified scheme reportedly has R-D performance comparable to prior Thomson approach.

6.1.1.1.4JVT-Y057-V ( Info) [M. Wien (Aachen U.)] Verif JVT-Y032 new results for Freeway sequence

This was a verification contribution relative to CE1 SVC on bit depth scalability for document JVT-Y032, reporting results for the Freeway sequence. The results of JVT-Y032 were reported to have been confirmed.
Summary of the verification task: The proponents provided the source code of the modified JSVM_8_12 software with BDS as distributed among the CE-participants, configuration files, simulation scripts, and bitstreams according to the specified test conditions.
For verification, the following steps were reported to have been performed:

• 
  The software provided by the proponents was successfully compiled under Linux and was used for the simulations.
  
• 
  The test sequences provided on the ftp site at Sharp and provided by the proponent were used.
  
• 
  Simulations were run using the simulation scripts provided by the proponents.
  
• 
  The provided bitstreams were successfully decoded.
  
• 
  The results were found to exactly match the results reported in the RD data provided with the streams. These results reportedly matched the results reported in contribution JVT-Y032. The results were reported in the accompanying Excel sheet.
  
• 
  An example configuration was encoded and was found to exactly match the bistream provided by the proponent.
  

Question: How carefully was this software studied? Answer: Verified by bitstream decoding; no thorough study of software.

6.1.1.1.5JVT-Y048 ( Prop 2.2/3.1) [Y. Ye, H. Chung, M. Karczewicz, I. S. Chong (Qualcomm)] Prop SVC bit depth scalability

In this contribution, an approach to bit depth scalability coding was proposed. This proposal relates to the difference between the HHI scheme and the Thomson scheme. Ways to “merge” the HHI scheme and the Thomson scheme are described. Simulations are only completed for the HHI framework. Simulations reportedly show that, for the Freeway sequence and the 12-bit sequences where the HHI scheme (prior to this meeting) underperformed the Thomson scheme, a modified method was reported to have performance on par with Thomson. Preliminary results reportedly indicate that a merge may also be applied to the Thomson scheme and used to bring up its performance, especially for the 10-bit sequences where it has underperformed the HHI scheme.
For the HHI scheme, the reason why it (previously) underperformed the Thomson method on Freeway sequence appears to bre because it disallows any temporal prediction to be carried out in the enhancement layer (10-bit MC). The same underperformance was reported to have been observed for the two 12-bit sequences (e.g. “Sunrise”). Lack of this prediction path can degrade coding efficiency when the additional bits carried in enhancement layer video are useful to form a more accurate temporal prediction. The JVT-Y048 proposal suggests to enable the BLskip mode to provide an alternative prediction path. It was suggested for the encoder to make an R-D decision between the existing IntraBL+inverse tone mapping mode, and the BLSkip mode. The encoder would then send a 1-bit flag per macroblock to indicate the selected mode. In the proposal, residual prediction is always turned off if the BLSkip mode is used.
Uses tone mapping and motion compensation (by introducing BL_skip mode). Requires multi-loop decoding and storage of both 8 bit and 10 bit results. For the 12-bit case, performance is not significantly more complex than simulcast.
Adds a requirement for the decoder to perform a high-bit-depth motion compensation.
Question: How much better than simulcast are the presented results? In presented results, there seems to be a high ratio of enhancement layer bits to base layer bits. Are the test conditions appropriate for the target application?
Question: Does it require multi-loop? Yes – requires full decode of base layer.
Question: Does it require storage of both the 8 bit and >8 bit layer decoded pictures? Yes.
Proponent suggestion is not immediate adoption of this scheme as-is, but rather further study of the issues.
A request was made for the latest results to be shown together, with comparison to simulcast, and to include Viper sequences.

6.1.1.1.6JVT-Y050-V ( Info 2.0/3.1) [Y. Gao and Y. Wu (Thomson)] CE1: Verif JVT-Y048 Qualcomm CE1 SVC prop

This was a verification contribution relating to CE1 on bit-depth scalability for document JVT-Y048.
The proponents of JVT-Y048 provided source code based on the bit-depth scalability reference software for JVT-X057, configuration files conforming to the test conditions defined in the CE1 description JVT-X301, simulation scripts, and experimental results.
For verification the following steps were reportedly performed:

• 
  The software provided by the proponents was compiled to generate encoder and decoder executables.
  
• 
  The test sequences provided on the bit-depth scalability AhG ftp site at Sharp were downloaded.
  
• 
  The reconstructed sequences (some other test sequences other than those for CE1) at the encoder and the decoder were matched.
  
• 
  Simulations were conducted using the simulation scripts provided by the proponents.
  
• 
  Results were reported in the provided Excel files.
  
• 
  The results were reported to exactly match the results reported in document JVT-Y048
  

Question: Was the software closely studied algorithmically or just run as provided by the other proponent? Answer: Just run. Verification done by only running software (no study).

6.1.1.1.7JVT-Y067 ( Prop 2.2/3.1) [S. Liu, A. Vetro (MERL)] CE1: SVC bit-depth scalability results

This document presented new results for the inter-layer prediction scheme for SVC bit-depth scalability proposed in JVT-X075. The new results were generated based on core experiment conditions defined in JVT-X301r1.doc. The merits of a block-based inter-layer prediction were also discussed.
The inter-layer prediction technique proposed in JVT-X075 applies a scale factor and offset value for each macroblock of each color channel. These values are predicted from spatially neighboring blocks and entropy coded. The proposed inverse tone mapping technique is independent of the forward tone mapping method and is reported to be able to cover a wide range of bit-depths. Also, the proposed method could reportedly be applied adaptively to different regions of interest, different frame pictures, and different color channels.
The key merits of the proposed approach in contrast to the global look-up table method reportedly include:

• 
  Adaptive to variety of tone mapping techniques, including locally adaptive tone mapping schemes, which could improve coding efficiency for such sequences.
  
• 
  Low delay encoding since inter-layer prediction decisions are made locally on a block basis without the need to determine the values of a LUT for an entire picture or group of pictures.
  
• 
  Reduced decoder complexity since implementing per pixel LUT is more costly than applying a block-based scale and offset.
  

Proponent recommendations:

• 
  Discuss the merits of block-based inter-layer prediction and consider adopting this approach as an alternative to the global LUT method currently defined in the JSVM.
  
• 
  Continue the CE to further evaluate inter-layer prediction techniques when different tone mappings are applied, e.g., to different regions or color components.
  

Some problems were evident for chroma channels – proponent is working on that.

Local adaptation by linear scaling plus offset function: Scaling factor, offset and prediction direction are encoded per macroblock (differentially) instead of tone mapping with LUT. Independent chroma prediction (independent of color format). It was claimed that the scheme could be benefical for larger bit depth ranges (8->12 bit) and for locally adaptive tone mapping schemes. Experimentally, some problems were found for chroma, approx. 2 dB loss (above 45 dB) compared to LUT method (bug? – preliminary results that improve shown in revision). For sequences with global tone mapping (Viper) LUT performs better overall. For 12-bit sequences with local tone mapping, LUT performs better below 45 dB (note: strange behaviour of LUT method at the rate point around 25 Mbps; LUT method still SPS based, not PPS based as presented in Y039). In the range of 50 dB and above, local adaptation by linear function provides gains of 0.5-0.7 dB for the two test sequences available.
Further study needed: More test material on 12 bit and diversity in common tone mapping algorithms needed. More evidence about relevance for application and potential profiles needed.
Extra overhead causes degradation of quality at low bit rates.
PSNR level is very high.
Remark: The results of the reference method shown in this contribution are for design status prior to the output of the last meeting – particularly not including PPS-level tone map adjustment capability.
Remark: How much better than simulcast? What are the bit rate ratios?
Remark: There is no 12-bit MC in this, so there is a complexity difference relative to 12-bit simulcast solution.
Proponent suggests merit of this approach in regard to local adaptation capability, such as region-of-interest application or other use of differing tone map use in different areas of picture.
Question: What is the encoder method to determine the parameters? Right now encoder is doing a search.
Remark: Decoder supporting both LUT and this approach would need two code paths and ability to switch between them, and syntax parsing capability.
Question: How common would we think a localized tone map use would be?
Remark: What are the application requirements? 12 bit 4:2:0?
Question: Is there a clear benefit for this spatial adaptivity capability?
Remark: In professional applications, increased fidelity has value for reasons other than visual viewing quality.
Proposed technique seems not beneficial for Viper sequences. Benefit seems small on the two other sequences.
Remark: Very strange behavior in some reference curves – e.g., decrease in objective fidelity with increase in bit rate.
We seem to need a better understanding of the needs and more test sequence data. Further study necessary.
Based on current results, no action to be taken.

6.1.1.1.8JVT-Y073-V ( Info) [A. Segall (Sharp)] Verif of JVT-Y067 MERL CE1 response

This document provides verification of Mitsubishi's response to CE1 on bit-depth scalability. Mitsubishi provided Sharp the software for JVT-Y067. Sharp reportedly carefully inspected the software, which included integrating Sharp's proposal into the software and performing a detailed comparison between JVT-Y067 and JVT-Y071. Sharp then compiled the software and generated results as specified in the CE. All results reportedly matched. JVT-Y067 was thereby reported to be verified.

6.1.1.1.9JVT-Y071 ( Prop 2.2) [A. Segall, Y. Su (Sharp)] CE1: Inter-layer prediction for SVC bit-depth scalability

This document provided a response to CE1: Inter-layer Prediction for Bit-Depth Scalable Coding. This response included results from JVT-X067, which proposed a modified inter-layer for bit-depth scalability. The scheme consists of a series of shifts and adds and is spatially varying.
This is conceptually very similar to JVT-Y067, and results were presented that included curves for each of these techniques. The Sharp method seems better at lower rates while MERL method seems better at higher rates.
An encoder-decoder mismatch problem occurred on one sequence.
A recommendation was made to further study 12-bit and localized tone mapping case.
See notes above in discussion of JVT-Y067.

6.1.1.1.10JVT-Y078-LV (Late Info 2.2/3.1) [S. Liu, A. Vetro (MERL)] Verif SVC JVT-Y071 CE1 results from Sharp

This document provided a cross-verification of the CE1 document JVT-Y071, proposed by Sharp. Both MERL and Sharp reportedly integrated their proposals, i.e. JVT-X075 for MERL and JVT-Y071 (JVT-X067r1) for Sharp, into the CE1 common software. The verification was based on the integrated software, and the bitstreams were generated according to the CE1 core experiment conditions. An accompanying Excel file contained simulation results with RD curves in comparison with the CE1 (LUT) results. The results of JVT-Y071 from Sharp were reported to have been confirmed.
Detailed source code review conducted. Results confirmed.

6.1.1.1.11JVT-Y072-L (Late Info) [A. Segall (Sharp)] Donation of tone mapped image sequences

In the previous Geneva meeting, there was an interest from the JVT for additional higher bit-depth data for testing bit-depth scalable coding tools. Sharp reported that it is currently able to donate two such sequences, and the purpose of this document was to document how the sequences were created.
The two donated sequences are currently stored on the Sharp FTP site that hosts the bit-depth and chroma format scalability AhG. The FTP is ftp.sharplabs.com, and the username and password are (bitdepth_ahg, 2bitd3pTh). The files are located in the directory 1080p/TwelveBit; stored in ZIP containers and titled “sunrise” and “library”.
The files were generated from high dynamic range video content that was originally stored in floating point format and in a linear RGB space. The 12-bit representation of the sequence was created by the following process:

• 
  RGB values normalized to the set [0, 1].
  
• 
  Normalized values were converted to YCC using the ITU-R BT.709 reference primaries.
  
• 
  Chroma planes subsampled by a factor of two in each dimension using the separable filter [-0.063 0.0 0.0299 0.0 -0.0831 0.0 0.3098 0.4994 0.3098 0.0 -0.0831 0.0 0.0299 0.0 -0.0063]
  
• 
  Resulting 4:2:0 data was quantized “linearly” to 12 bits of precision with a rounding operation.
  

The output was then stored in a YUV file container as 16-bit data, with the 12 bit data stored in the least significant bits.
Creation of the corresponding 8-bit data was accomplished by operating on the original, floating point image. It was created using the following process:

• 
  Image tone mapped to 8-bits using a spatially varying operator described in documents referenced in the contribution. Here, the output corresponds to the LCD image of an HDR display and is stored as 8-bit R'G'B'.
  
• 
  Tone mapped output converted to YCC using the ITU-R BT.709 reference primaries.
  
• 
  Chroma planes subsampled by a factor of two in each dimension using the separable filter [-0.063 0.0 0.0299 0.0 -0.0831 0.0 0.3098 0.4994 0.3098 0.0 -0.0831 0.0 0.0299 0.0 -0.0063]
  
• 
  Resulting 4:2:0 data was quantized “linearly” to 8 bits of precision with a rounding operation.
  

Note: Data are 12-bit 4:2:0. Remark: Is this realistic?

Thanks were expressed to the contributors. A copyright statement was included in the zip container.

6.1.1.1.12JVT-Y080 ( Prop 2.2/3.1) [Y. Gao, Y. Wu (Thomson)] CE1: SVC combined spatial and bit-depth scalability

This contribution presented simulation results of Thomson’s proposed technical solution to bit-depth scalability JVT-X051. The software integration and the test conditions in the performance test were subject to JVT-X301. In particular, simulations results of combined spatial and bit-depth scalability were provided. Complete simulation results including the PSNR-bitrate points for Y, Cb, and Cr channels were reported.
This contribution proposes a technical solution to bit-depth scalability that was reportedly conforming to the current SVC standard (exact meaning?). The inter-layer bit-depth prediction was realized by a left bit-shift of each channel (Y, Cb and Cr) of the base layer MB to predict the co-located enhancement layer MB (which can be supported by the current inverse tone map design). This process was suggested to be called “upsampling process for bit depth prediction” and is proposed to be invoked when bit depth scalability is enabled. With the proposed solution, there was reported to be no new syntax elements needed to support the proposed solution.
Uses bit shift/truncation for the mapping between 8 and 10 bit (no tone mapping); unlike the current JSVM LUT tone mapping, this uses motion compensation in the enhancement layer.
Motion compensation needed for spatial scalability.
Comparable to case 3 of JVT-Y039 – both parties were asked to report back with a comparison of their results – see BoG report JVT-Y088 and related notes. Note: 3-layer approach may be less complex because higher spatial resolution MC is performed with 8 bit.
The proposal uses MC at the enhancement layer, unlike the current reference adopted method, along with inter-layer prediction with bit depth expansion.
Question: Comparison to current reference method (the one with 8-bit MC at lower layer only)?
Without the spatial scalability aspect, we have the same situation as at last meeting, basically. Except for the “freeway” sequence, the reference method does better. With the new 3-layer rewrite approach JVT-Y032, the compression results seem about the same for the two approaches.
How about a 3-layer approach – use a spatial scalability enhancement layer that is then enhanced with bit depth scalability. JVT-Y039 reportedly includes such results as its “configuration 3”, which seems directly comparable to what is reported in this contribution.
How do the results compare? Seem roughly comparable at first glance, and simulcast relationship should be considered for such a comparison. Interested parties were asked to look at these issues off-line. See BoG report JVT-Y088 and related notes.

6.1.1.1.13JVT-Y081 ( Prop 2.2/3.1) [Y. Wu, Y. Gao (Thomson)] CE1: SVC study on inter-layer prediction: bit-depth scalability

This contribution presented further study on inter-layer prediction within the framework of bit-depth scalable coding proposed in JVT-X051, which uses motion compensation at the enhancement layer bit depth.
It focused on enhancing the JVT-X051 method by employing inter-layer prediction techniques other than linear scaling.
As an example, the LUT based inverse tone mapping algorithm (proposed in JVT-X057) is applied to the enhancement layer macroblock of which the co-located base layer macroblock is intra-coded. Experimental results comparing the LUT based inverse tone mapping technique and the linear scaling in inter-layer prediction were provided. The contribution also provides some experimental results on the 12-bit test sequences by enabling both adaptive inter-layer prediction and the proposed LUT based inverse tone mapping in INTRA_BL mode.
Uses lookup table for 8 to 10 bit mapping, and also MC in the 10-bit layer. Shows improvement relative to the previous proposal (with bit shift = linear mapping).
Remark: Do we have results of comparing the proposed scheme relative to the reference scheme that does not use MC with high bit depth? Not provided in contribution.
Interested parties were asked to look at these issues off-line.
Comparison with results from JVT-Y032 and JVT-Y039 was asked to be made and reported. It was suggested to review then to find out how much is gained by MC in the 10 bit layer.
See BoG report JVT-Y088 and related notes

6.1.1.1.14JVT-Y049-V ( Info) [I. S. Chong, Y. Ye, M. Karczewicz (Qualcomm)] CE1: Verif JVT-Y080 and JVT-Y081 Thomson CE1 SVC props

This document provided a verification of JVT-Y080 and JVT-Y081, which were Thomson's CE1 proposals. Thomson reportedly provided Qualcomm with the source files for JVT-Y080 and JVT-Y081 which we reportedly inspected and compiled. And Thomson provided config files and Excel spreadsheets. Sequences were obtained from the CE1 ftp site. Qualcomm then generated data points according to the testing conditions in CE1 and compared with the Thomson spreadsheets. The results were reported to match.
Question: How close was the source code inspection?
Participants of CE were asked to provide comparisons of their newest results in common figures; for 12-bit case, it was requested to clarify what the useful rates and application scenarios are, what typical simulcast rates would be, and where we stand as compared to simulcast.
See BoG report JVT-Y088 and related notes

6.1.1.1.15JVT-Y088-B [A. Segall] BoG report on bit depth and chroma format scalability

Results of work on bit depth and chroma format scalability were presented as JVT-Y088, and are summarized as follows:

• 
  What’s the application for BDS?
  
  • 
    Support professional/higher end displays and “traditional” display in single bit-stream
    
  • 
    Provide “premium” high bit-depth content
    
  • 
    Observation from Thomson
    
    • 
      Saw visual difference for computer generated content at 10-bit vs. 8-bit
      
    • 
      Did not see visible difference between 10-8 for film
      
  • 
    We must be able to see differences (subjective tests)
    
  • 
    More information (liaison statements to BDA to HD-DVD)
    
  • 
    Blu-ray/HD-DVD/Broadcast
    
    • 
      8-bit decoders
      
    • 
      HDMI 1.3 (30, 36/48)
      
    • 
      Must fit…
      
  • 
    Production/post-production
    
    • 
      Needs: reduce storage costs of managing multiple versions of sequences
      
    • 
      Support different resolutions (HD to 4k-by-2k)
      
  • 
    Quantify needs …
    
• 
  Consumer Applications: Blu-ray/HDDVD
  
  • 
    Two layer system with same spatial resolution
    
  • 
    Resolution: HD (1080p24)
    
  • 
    Total bit-rate not to exceed 29Mbps/40Mbps
    
  • 
    No additional constraint on baselayer bitrate (with exception of video bit-rate constraint)
    
  • 
    No sacrifice of visual quality of base layer
    
  • 
    Enhance layer
    
    • 
      Higher bit-depth
      
    • 
      Higher chroma format? 4:2:2 or 4:4:4?
      
    • 
      Wider color gamut: YCC to xvYCC support
      
    • 
      Percentage of bits
      
      • 
        4:2:0 both layer: target 25%
        
      • 
        4:2:0 to 4:2:2:/4:4:4: target ? (but probably something higher than 25%; perhaps 40%)
        
• 
  Recommendations from BoG:
  
  • 
    Continue CE and AhG
    
  • 
    Liaison letters to BDA and HD-DVD forum
    
  • 
    CE sequences to include:
    
    • 
      Source type:
      
      • 
        Viper (Freeway, waves, plane, CR and staples)
        
      • 
        No SVT
        
      • 
        sunset ?
        
      • 
        NASA, CG (further study)
        
      • 
        Desire to test total of 8
        
    • 
      Stay with 1080p
      
    • 
      Try to test both 10/12 bit data
      
    • 
      TMO (linear: 1, global: 4, spatial: 2)
      
  • 
    Mandate of AhG to include 4:2:0 to 4:2:2/4:4:4 scenario
    
• 
  Also desire to test 4:2:0->4:4:4
  
  • 
    For all sequences: generate 4:4:4, convert to 4:2:2, convert to 4:2:0, convert to 8-bit
    
  • 
    New mandate for AhG
    
• 
  SVT data may be redundant … consider 1-2 sequences
  
• 
  Viper data: capital records, night, staples show very similar visual characteristics; suggest using only 1-2 (choose the one with scene cut)
  

The results of prior experiments were consolidated into accompanying spreadsheets and reviewed – however, no proponent seems to have provided full results for all test sequences.

Remark: Combination of bit depth and spatial scalability does not appear useful.
Comment: Things might look different with some new results – perhaps 1 dB relative to simulcast.
Remark: There were differences between simulcast scenarios input to this meeting.
Suggestion to redefine simulcast comparison in terms of focusing on quality differences at equal total bit rate…
The test data available seems inadequate.
Various factors were discussed.
Visual evaluations seem necessary to be able to make conclusions about designs.
The effort so far has been basically an exploration area – further input would be needed to clarify the market needs.
Our situation: Further clarification of the need and usefulness of the technology seems necessary before proceeding with a standardization project on bit depth or chroma format scalability.