International organisation for standardisation organisation internationale de normalisation

No.	Title	TBP
Available
12549	Request for Subdivision: ISO/IEC 15938-13 Compact Descriptors for Visual Search	N	12/02/10
12550	Test Model 1: Compact Descriptors for Visual Search	N	12/02/20
12551	Core Experiments in Compact Descriptors for Visual Search	N	12/02/10

7Internet and web video coding

A small editors' group worked in a side activity to produce an improved WD version for Web Video Coding. Unlike what was anticipated in the previous timeline plans, no progression to CD was made yet.

Documents M23781, M23786 and M23789 were reviewed in the Video group for information, without action.
A break-out group on Internet Video Coding (IVC) met several days throughout the week. Since the last meeting, the performance of the test model (respectively its software implementation) had been improved over the previous version due to bug fixing. This reportedly resulted in approximately 8.2% bit rate reduction (for the same PSNR on average) for IBBP picture sequence structures, and 6.7% for IPPP coding. Beyond that, only a few new tools had been integrated, e.g. quarter-pixel motion compensation using bilinear interpolation filters. It was however evident that there was still incorrect behaviour in several parts of the encoder (e.g. rate-distortion optimization) which would prevent the codec to achieve better performance. The encoder still requires more than 20% on average higher bit rate than AVC Constrained Baseline Profile for same quality (when measured in terms of PSNR). The main discussion item during the week in the breakout group was therefore to find and fix more problems with the algorithm, and develop a workplan in preparation for the next meeting, and also to define Core Experiments. Whereas the former effort did not lead to substantial further improvement, a work plan was defined as follows:

1. 
   Verification of test model software:
   
   1. 
      Write test code to perform forward and inverse transforms for each size, and verify the result is identical.
      
   2. 
      Substitute the JM motion estimator, and compare results.
      
   3. 
      Perform more tests by substituting the JM sub-pixel interpolator, and compare results.
      
2. 
   Analysis of test model software:
   
   1. 
      Continue macroblock (MB)-level comparison of RFM and JM:
      
      1. 
         Improve RFM software analysis to include motion vector (MV) data: values and residual bits.
         
      2. 
         Ditto for coefficient residuals.
         
      3. 
         Check block size and motion type.
         
      4. 
         Check skip at block level and MB level.
         
   2. 
      Analyze many MB for different test sequences at different rates to look for major differences between RFM and JM:
      
      1. 
         When anomalies are found, analyze inside the MB and block-level code to determine their origin.
         
   3. 
      Review RDO process and code:
      
      1. 
         Check the algorithm and its code.
         
      2. 
         Run a series of experiments with different lambda values to look for any issues.
         
   4. 
      Verify arithmetic coder operation:
      
      1. 
         Write a test program to collect statistics on the core engine performance
         
         1. 
            For coeff. residuals and MVs, gather statistics on the % of time the prediction is correct for each bit in the value.
            
         2. 
            Modify the context tables to see the impact; and tune the tables based on the observations.
            
         3. 
            Similarly investigate the behaviour for MB mode information, block size, and motion direction.
            
      2. 
         Study context accuracy from one value to the next – and analyze whether a syntax element such as block size is accurately predicted from one block or MB to the next.
         
3. 
   Tuning of test model software:
   
   1. 
      Encoder issue study:
      
      1. 
         Lower QP for I pictures (since there is no non-DC intra prediction).
         
      2. 
         Promote skip mode.
         
      3. 
         Use SATD instead of SAD for ME.
         
   2. 
      Arithmetic coder operation:
      
      1. 
         Study best path for context at left edge of picture/slice.
         
      2. 
         Study best path from one block to the next, especially when block size changes.
         
4. 
   Fix error in chroma transform size.
   
5. 
   Update test model document based on results of testing.
   
6. 
   Produce a WD.
   

Core experiments were defined on integer DCT, quad-tree based variable block size coding, improvement of entropy coding, motion compensation accuracy, intra and inter prediction improvements, and encoder optimization.

AHGs were appointed on Web Video Coding and Internet Video Coding (N12575 and N12576, as recorded in N12453) to conduct further work on development of these approaches until the 100th meeting.
Document(s) reviewed:

m22880	Ad hoc on Internet Video Coding	Siwei Ma, Yun He
m22879	Ad hoc on Web Video Coding	Krasimir Kolarov, David Benham, Gaëlle Martin-Cocher, Jesus Sampedro, Dave Singer, Thomas Wiegand
m23557	JNB comments on Web Video Coding and Internet Video Coding	Japan National Body
m23921	CE Report on Internet Video Coding	Siwei Ma, Xin Zhao, Yunfei Wang, Jianwen Chen, Yun He
m23778	Updated Web Video Coding WD	David Benham, Gaëlle Martin-Cocher, Eric Pearson, Jesus Sampedro, David Singer, Thomas Wiegand
m23781	Known Patents on MPEG WebVC Proposal	Rob Glidden, Cliff Reader
m23786	Market Developments Relating to MPEG Royalty-Free Standardization	Rob Glidden, Cliff Reader, Gerard Fernando
m23789	Additional Known Patents on MPEG WebVC Proposal	Rob Glidden, Cliff Reader, Gerard Fernando

Document(s) approved: