JVET-AL0010 JVET AHG report: Encoding algorithm optimization (AHG10) [K. Andersson, P. de Lagrange, A. Duenas (co-chairs), T. Ikai, T. Solovyev, A. Tourapis (vice chairs)]

6 contributions and one cross-check document were identified relating to AHG10 and summarized in the following sections.

JVET-AL0055: AhG10: On constrained encoder complexity configuration of VTM

The document describes three alternative configurations for the VTM to achieve additional complexity performance trade-offs. A preferred variant shows an impact of 37%/46%/44%/43% to the encoder runtime with a bd rate impact of 4.47%/1.49%/4.17%/4.34% relatively to the default VTM 23.5 RA, AI, LDB, and LDP configurations respectively. The details of the preferred configuration are shown below:

Less extensive partitioning search:

--MaxMTTHierarchyDepth=1

--MaxMTTHierarchyDepthISliceL=2

--MaxMTTHierarchyDepthISliceC=2

Implicit MTS instead of explicit one:

--MTS=0

--MTSImplicit=1

Lower number of merge candidates:

--MaxNumMergeCand=5

A cross-check of this proposal is available in JVET-AL0263.

JVET-AL0074: AHG10: Teleconference on encoder optimization

The document summarizes discussions held during a teleconference on February 24, mainly related to multi-layer spatial scalability, but also to multi-layer multi-view, multi-layer content layering, and reduced complexity performance trade-offs.

JVET-AL0114: AHG10: On intra mode selection

The document proposes modifications to the intra mode selection of the ECM specifically to reduce the matrix-based position-dependent intra prediction (MPDIP) mode tests. The following results are currently included in this contribution.

All-Intra: {-0.01%, 0.04%, 0.02%, EncT: 99.3%, DecT: 99.8%}

Random-Access: {} (Note: currently missing)

JVET-AL0207: AHG10/AHG17: On matching target bitrate for subjective quality evaluation

The document proposes an alternative approach for matching the target bitrate for subjective evaluations. Instead of increasing the base QP by one at a given frame, which may give an abrupt change in quality after the QP change, it is proposed to start from the base QP closest to the target bitrate and to adjust the lambda to correspond to a fractional QP that is in-between the two closest QPs. In this way it is claimed that a more equal quality can be reached for all frames rather than having a change of quality for frames after a change of the base QP.

The document proposes to add this functionality in the VTM reference software and suggests using the method for matching a target bitrate in the context of subjective quality evaluation.

JVET-AL0213: AHG7: On common test conditions for fixed-QP objective rate-distortion testing of "low delay" encoding

The document requests JVET to consider modifying the method used for routine objective rate-distortion testing of “low-delay” configurations for high-resolution video content to combat with the encoding of a long series of frames in a sequential manner. The suggested alternative is to divide a test sequence into closed-GOP segments that are coded independently of each other and to measure only the bit rate and distortion for the non-intra frames. A hypothetical alternative approach of using temporal sublayers is also suggested to be a possible way to reduce encoding times. In such a scheme, the lowest temporal layer would be coded first, followed by parallel encoding of the higher temporal sublayers, with all frames using only forward temporal prediction.

The intent of this document is to stimulate the discussion and study of the potential approaches for reducing encoding times for such tests.

JVET-AL0233: Information on low complexity encoding experiments

The document claims that the ability of supporting a low complexity encoder with reasonable compression efficiency is an essential functionality for a next generation video coding specification design. Experiments related to the ECM have shown that changing the encoder’s configuration can have a large impact on the encoder complexity especially in later versions of the ECM. Some of the operating points seem to achieve a better RD-complexity trade offs than the ECM anchor points, and some operating points seem to might be able to reach the desired complexity in encoding time. The latest version of the ECM, ECM-16.0 seems unable to reach the desired complexity in both the encoder and the decoder. A significant modification of the design and the software may be required to reduce decoder complexity such that it can reach the desired decoder complexity.

JVET-AL0245: [AHG7][AHG17] Further consideration of common test condition

This document further explores constrained test conditions by analyzing experiment results on ECM-16.1 under various MTT depth settings. It recommends configuring the encoder setting during the development of the next standard to approximately 2x to 4.5x of the encoding time of the VTM, and by changing the MTT depth parameters to the following (about 3x faster with around 4% BD-rate loss):

--MaxMTTHierarchyDepth=2 --MaxMTTHierarchyDepthISliceL=2

--MaxMTTHierarchyDepthISliceC=2 --MaxMTTHierarchyDepthByTid=221111

1. Recommendation

The AHG recommends that the related input contributions are reviewed, and to further continue the study of encoding algorithm optimizations in JVET.