Joint Collaborative Team on Video Coding (jct-vc)

6.3BoGs

JCTVC-T0229 BoG report on test sequences for SCC [H. Yu (BoG coordinator)]

(Consideration of this BoG report was chaired by GJS on Sunday 02-15 p.m.)

The BoG met on the morning of Feb. 15, 2015. The purpose of this BoG was to make recommendations on the test sequences to be used after this meeting. About 15 experts attended the discussion.

According to the relevant comments made in the discussion of T0006 in the meeting notes, the BoG suggested to add more 4:2:0 sequences to the CTC. Specifically, the BoG suggested converting the following CTC sequences into 4:2:0 format:

• 
  6 TGM (i.e. all except SlideShow)
  
• 
  3 mixed content
  
• 
  1 animation
  

In addition, the BoG suggested using the following sequences from Class F:

• 
  SlideShow
  
• 
  ChinaSpeed
  

A table showing all the selected test sequences and their classifications was presented in the BoG report.

The BoG suggested using a simple filter to do the conversion. Qualcomm offered a 2-tap filter [0.5, 0.5] for the initial conversion.

In the subsequent JCT-VC discussion it was noted that properly the horizontal filter should be an odd length filter, and agreed that [0.25, 0.50, 0.25] will be used horizontally (with adjustment for edges of the picture as needed).

Sample converted sequences will be available for review during this meeting.

The BoG suggested that it would be good to have new test material with non-alphabetical texts, e.g. Chinese text in a browser. However since we do not have such sequences at present time, we cannot add such a sequence yet. The BoG suggested to encourage participants to bring new sequences in the next meeting.

The recommendations of the BoG were agreed by the JCT-VC (with the noted revision of the filtering method to be applied).

JCTVC-T0230 BoG report on IBC memory bandwidth reduction [K. Rapaka, M. Zhou (BoG coordinators)]

(Consideration of this BoG report was chaired by GJS on Sunday 02-15 p.m.)

This document provides a report of the BoG on intra block copy (IBC) memory bandwidth reduction.

Relevant proposals were identified as T0045, T0051, and S0145.

In the BoG, it was suggested to investigate the IBC memory bandwidth analysis under the full frame search area configuration. The bandwidth analysis includes the worst-case and average case memory bandwidth.

Methods recommended to be investigated by the BoG included the following:

1. 
   JCTVC-T0045/ JCTVC-T0051: signal the CTBs which are available for IBC prediction. For these CTBs, the filtering is disabled.
   
2. 
   Combination Test: The encoder method of JCTVC-T0045 is added over JCTVC-T0051
   
3. 
   Signal the CTBs which are available for IBC prediction without disabling filtering operation (JCTVC-S0145).
   
4. 
   Restrict that the any reference pixel for Intra BC cannot be from bi-predicted blocks. This method will have same worst case bandwidth as HEVC.
   
5. 
   Sequence and picture level control: – The filtering is disabled sequence level and adaptively at picture level.
   
6. 
   Under CTC, further reduce the number of inter reference pictures for RA/LD-B by 1.
   

The IBC search range configurations recommended to be tested by the BoG are specified shown in the following table:

Configuration	8x4	4x8/8x8/16x16/32x32
Config 1	2x1-CTU	Full-frame
Config 2	4x1-CTU	4x1-CTU

In the review of the BoG report, it was remarked that this table may just be reflecting the ordinary planned CTC encoding method, and thus the table may not be needed – and further remarked that we should just use the ordinary CTC in any case.

AAn additional test configuration was recommended by the BoG: using a CTU size of 16x16 and local search area size in terms of samples that is kept the same as config 1 (128x64).

Expert subjective analysis was suggested by the BoG to be performed for the methods that change filtering operations (Deblock/SAO).

It was suggested by the BoG to use higher QP (in the range of QP=37) for the visual testing purposes.

It was also suggested by the BoG to test under LD-P configuration for the visual testing purposes, in addition to the usual LD-B configuration.

It was agreed in the BoG that the methods 1, 2, and 4 are reasonable only with use of local cache.

The BoG recommended that there be a CE on the topic.

In the JCT-VC discussion of the BoG report, the rationale for reducing the number of reference pictures was questioned somewhat, and it was not entirely clear what is to be tested for item 5 and its need was somewhat questioned. (Items 5 and 6 were suggested by non-proponents, and the proponents indicated some doubts about those aspects.)

The BoG outcome will be taken into account in the CE plan.

6.4Summary of normative decisions

The following is a summary of the normative decisions made at the meeting for the HEVC specification draft text (to be included in the draft text of the screen content coding extensions):

• 
  Palette mode coding:
  
  • 
    As an editorial correction, add missing inference rule for palette_escape_val_present_flag semantics. See notes for T0183.
    
  • 
    Palette prediction modification in relation to tiles and slices, where a PPS-level table is used to initialize the palette predictor. See notes for T0048.
    
  • 
    Double the maximum allowed palette size and palette predictor size to 64 (this may not be a normative change, depending on whether this is just a change of agreed test conditions or a change of a normative limit expressed in the text). See notes for T0052.
    
  • 
    Disallow palette mode for 64x64 CUs. See notes for "method 2" of T0058.
    
  • 
    Modification of escape colour coding method in regard to fixing the maximum value cMax specified for truncated binary code for escape colours and the formula for inverse quantization of escape-coded colours (but specify that when cu_transquant_bypass_flag is equal to 1, the cMax is (1 << bit_depth) − 1). See notes for T0118.
    
  • 
    Remove the palette sharing flag and its context, and bypass code the binarization of the number of new palette entries. See notes for T0064.
    
  • 
    Grouping together the palette indices and sending them before sending the run type and run lengths. See notes for T0065.
    
  • 
    Grouping together the escape pixel values at the end of the palette coded data for the CU. See notes for T0076.
    
  • 
    Remove one context for run_type coding. See notes for T0078.
    
  • 
    Change the coding for palette_num_signalled_entries from unary coding to zero-order Golomb coding. See notes for T0063.
    
  • 
    Palette coding modifications for non-4:4:4 chroma format support. See notes for T0072, T0109, and T0120.
    
  • 
    In the SPS syntax, code the maximum palette table predictor size as a delta increase relative to the maximum palette table size. See notes for T0134.
    
• 
  Intra block copy:
  
  • 
    Unification of intra block copy with inter mode signalling and design aspects. See notes for T0227.
    
• 
  Intra MPM derivation
  
  • 
    Add a missing initialization value of DC as needed when deriving the MPM for intra prediction. See notes for T0049.
    
• 
  Cross-component prediction and adaptive colour transform
  
  • 
    Specify a normative constraint that limits inputs to the inverse CCP to the range of the quantized coefficients CoeffMinY/C to CoeffMaxY/C when ACT is not used. This is a bug fix to RExt as well as a decision applying to SCC. See notes for T0003, T0007, and T0132.
    
  • 
    Add clipping at the input of ACT in the decoder, with the same range as is specified for the normative constraint on the input to the inverse CCP – i.e., clipping the output of the inverse CCP to the range of the quantized coefficients CoeffMinY/C to CoeffMaxY/C. The clip is to be applied before the left shift of the chroma by one bit position for the input of the inverse CCP. See notes for T0132.
    
  • 
    Modified QP offset signalling method for switching QP offset values between RGB colour space and YCoCg colour space residuals when adaptive cross-component transform is used. See notes for T0140.
    
• 
  Adaptive motion vector resolution
  
  • 
    When integer motion operation is selected, clip the range of horizontal and vertical components of MV values to a 16-bit range after the left shift that causes the value to become an integer. See notes for T0059.
    
  • 
    In the SPS syntax, change the adaptive_mv_resolution_enabled_flag to a two-bit mode indicator for motion vector resolution control, to avoid needing to send the use_integer_mv_flag in every slice header when it will always be 1. See notes for T0069.
    
• 
  Supplemental enhancement information:
  
  • 
    Add coded region completion SEI message with multi-layer adjustments. See notes for T0047. (This decision was inadvertently omitted in the post-meeting editing of the output text, as was another SEI message, the DRAP SEI message, which was adopted at the Strasbourg meeting based on document JCTVC-S0095.)
    
  • 
    Add content light level SEI message. See notes for T0101.
    

For further detail, see notes in other sections. The above list is only provided as a best-effort summary.