JVET-F0002 JVET AHG report: JEM algorithm description editing (AHG2) [J. Chen, E. Alshina, J. Boyce]

This document reports the work of the JVET ad hoc group on JEM algorithm description editing (AHG2) between the 5th JVET meeting at Geneva, Switzerland (12–20 January 2017) and the 6th meeting in Hobart, AU (31 March – 7 April 2017).

During the editing period, on top of the JVET-D1001 Algorithm Description of Joint Exploration Test Model 4, the editors worked on the following aspects to produce the final version of JVET-E1001 Algorithm Description of Joint Exploration Test Model 5.

• Integrating adoptions of the 5th JVET meeting that changed the encoding or decoding process,
  • JVET-E0052, decoder-side motion vector refinement based on bilateral template matching with integer pel step search only (8 positions around the start position with integer pel MV offset)
  • JVET-E0060, test 3 case, FRUC with additional candidates
  • JVET-E0076, MVD coding in unit of four luma sample precision
  • JVET-E0062, multiple Direct Modes for chroma intra coding: the total number of chroma intra modes is kept as 6 (unchanged), the list construction of chroma mode is modified with the first six as proposed
  • JVET-E0077, enhanced Cross-component Linear Model Intra-prediction, includes
    • Multiple-model Linear Model intra prediction
    • Multiple-filter Linear Model intra prediction
  • JVET-E0104, ALF parameters temporal prediction with temporal scalability
  • JVET-E0023, improved fast algorithm test case B: skip depth is set equal to 2 always for LDB and LDP, skip depth is set equal to 2 for highest temporal layer in RA, and is set equal to 3 for other temporal layers
• Overall text refinement and quality improvement (esp. FRUC description enhancement with details)
• Fix of text and software bug related to transformed coefficients zero-out for large transform (Ticket#44)
  • Note: This aspect is not in the output of the previous meeting – planned for incorporation into the next JEM description

Currently the document contains the algorithm description as well as encoding logic description for all new coding features in JEM5.0 beyond HEVC. Compared to HEVC, the following new coding features are included in JEM5 (where boldface indicates aspects that had technical changes incorporated as the result of the previous meeting):

1. Block structure
   1. Larger CTUs (software supports 256×256, CTC use 128×128)
   2. Quadtree plus binary tree (QTBT) block structure
2. Intra prediction
   1. 65 intra prediction directions with improved intra mode coding
   2. 4-tap interpolation filter for intra prediction
   3. Boundary filter applied to other directions in addition to horizontal and vertical ones
   4. Cross-component linear model (CCLM) prediction
   5. Position dependent intra prediction combination (PDPC)
   6. Adaptive reference sample smoothing
3. Inter prediction
   1. Sub-PU level motion vector prediction
   2. Locally adaptive motion vector resolution (AMVR)
   3. 1/16 pel motion vector storage accuracy
   4. Overlapped block motion compensation (OBMC)
   5. Local illumination compensation (LIC)
   6. Affine motion prediction
   7. Pattern matched motion vector derivation
   8. Bi-directional optical flow (BIO)
   9. Decoder-side motion vector refinement
4. Transform
   1. Larger transform block sizes (with zeroing of high frequencies)
   2. Explicit multiple core transform
   3. Mode dependent non-separable secondary transforms
   4. Signal dependent transform (SDT, disabled by default)
5. Loop fillter
   1. Adaptive loop filter (ALF)
   2. Content adaptive clipping
6. Enhanced CABAC design
   1. Context model selection for transform coefficient levels
   2. Multi-hypothesis probability estimation
   3. Initialization for context models

Among all of them, the decoder-side motion vector refinement (DMVR) method was added at the 5^th JVET meeting. The AMVR method is modified by adding MVD coding in units of four luma samples, in addition to quarter luma sample and integer luma sample. The CCLM method is modified by adding multiple-model Linear Model intra prediction and multiple-filter Linear Model intra prediction.

The AHG recommended to:

• Continue to edit the algorithm description of the Joint Exploration Model document to ensure that all agreed elements of the JEM are described
• Continue to improve the editorial quality of the algorithm description of Joint Exploration Model document and address issues relating to mismatches between software and text.
• Identify and improve the algorithm description for critically important parts of JEM design for better understanding of complexity.

It was asked whether we need to maintain proper operation of the JEM with QTBT disabled. It was agreed that this remains a highly desirable capability, but has lower priority than other matters.

It was commented that removing this capability would make the code much cleaner and substantially reduce the size of the codebase.

Another participant suggested that pseudo-profiling (i.e., supporting on/off switches in syntax) rather than the use of a macro can make it easier to maintain proper functioning. However, an AHG co-chair said that this is difficult to do for aspects that affect basic structure, like QTBT.

It was commented that some bug fixes are found in the software with macros and that in such cases the macro should just be removed.

It was commented that we still need to integrate screen content coding tools and enable non-4:2:0 operation.

Decision (CTC): It was agreed that in the CTC when Class F is tested, the SCM (with SCC tools enabled) should be substituted for the HM anchor.