JVET-AE0010 JVET AHG report: Encoding algorithm optimization (AHG10) [P. de Lagrange, A. Duenas, R. Sjöberg, A. Tourapis (AHG chairs)]
Related contributions
A total of 5 contributions, not including cross-checks, were identified relating to AHG10, and summarized in the following sections.
JVET-AE0104 - AHG10: GOP-based RPR encoder control using parallel resolution encoding
This contribution proposes multi-pass GOP-based adaptive resolution across full-resolution and 4/5, 2/3, and ½ reduced resolutions, using a rate-distortion cost based on squared error (=PSNR). The reported BD-rate difference in RA class A (using CTC QPs) are about -1.5% when compared to regular full-resolutions, and around -1% when compared to the current single-pass GOP-based RPR technique implemented in the VTM. When adding a lower quality point on top of the CTC QPs (QP=42 for UHD), gain increases compared to regular coding, but decreases compared to current GOP-based RPR. Gains are generally lower for lower resolutions and for LD and AI configurations.
Low delay
JVET-AE0114 - AHG10: Low-Delay configuration improvements
This contribution proposes changes to the low-delay configurations regarding pictures and QPs for improved coding efficiency. In addition to a -3 QP offset to the first intra picture (instead of -1), it is proposed to introduce a multi-level QP structure with a periodicity of 16 pictures, plus a picture with a -2 offset every 64 pictures. Reference picture list is changed to take advantage of higher quality reference pictures. A BD-rate difference of around -6% is reported.
Bug fixes, more encoder control
JVET-AE0111 - Additional conformance tests of spatial scalability for multilayer coding
Besides proposing an additional conformance bitstream, this contribution reports about fixing a bug in reference picture list when combining scalability and low-delay configuration: first picture of enhancement layer is intra-coded (not using the base layer as a reference), and for the remaining pictures, intra-layer prediction is removed when inter-layer referencing is activated; the proposed fix is in MR !2633.
JVET-AE0122 - AHG10: GOP-based RPR and scalable coding
This contribution proposes a bug fix and more control over the encoder when combining RPR and multi-layer. One use case is to restrict reduced resolution coding to the base layer. A marginal BD-rate gain is reported when testing the proposed changes using 1.5x scalability and GOP-based RPR for the base layer, over the test sequences currently considered for scalability visual testing.
JVET-AE0181 - Scaling window support for VTM
This contribution reports about adding control of the scaling window to the VTM, with additional configuration parameters. It also reports about an experiment using dual-layer coding with the enhancement layer containing a scaled-down, windowed version of the base layer. When coding without scaling window, affine coding mode is mostly used, while merge mode is used when using scaling window. A lower rate with a higher PSNR is obtained.
Recommendation
The AHG recommends that the related input contributions are reviewed and to further continue the study of encoding algorithm optimizations in JVET.