Page 102 - ITU Journal, ICT Discoveries, Volume 3, No. 1, June 2020 Special issue: The future of video and immersive media

ITU Journal: ICT Discoveries, Vol. 3(1), June 2020

release in March 2018. The libaom software Table 3 – Broadcast: VMAF BD-rate versus HM
corresponding to this version was used in [12] with Broadcast
two-pass, constant quality (cq) but with cpu-used=1 VMAF VTM8 libaom ETM4.1
(not the best quality). The BD-rate gain over HM
was 10%. The same year [17] measured 17% gain UHD −42.8% −19.6% −33.9%
in the same configuration. In 2019 [18] tested the HD −40.8% −21.0% −30.7%
current libaom software with one-pass only on HD WVGA −29.2% −12.5% −20.3%
and UHD format resulting in the same performance WQVGA −30.7% −19.5% −22.1%
as HM.
Overall −36.9% −18.5% −27.7%
The overall gain reported in Table 1 of 14.7% for Diff vs VTM 0.0% 18.4% 9.2%
libaom over HM is lower than the gain of 17%
reported in [17], but in the same order of magnitude, Table 4 – Broadcast: MS-SSIM BD-rate versus HM
in very similar testing conditions. The difference Broadcast
could come from the disabling of quantization MS-SSIM VTM8 libaom ETM4.1
parameter variation within a picture (deltaq=0).
UHD −39.5% −14.5% −29.9%
Table 1 – Broadcast:PSNRYUV BD-rate versus HM HD −32.9% −10.7% −23.6%
Broadcast WVGA −28.1% −7.2% −17.2%
PSNRYUV VTM8 libaom ETM4.1 WQVGA −25.0% −6.1% −15.5%
UHD −41.9% −18.0% −28.4% Overall −34.3% −11.3% −24.5%
HD −39.0% −16.3% −21.0% Diff vs VTM 0.0% 23.0% 9.8%
WVGA −30.8% −11.4% −17.6%
WQVGA −28.4% −11.1% −16.3% 5.1.2 Streaming

Overall −36.0% −14.7% −21.6% In the Streaming scenario, Table 5 shows the VTM
Diff vs VTM 0% 21.3% 14.4% still performs better than the other solutions with
approximatively the same figure (14.7%) against
Objective metrics in Table 2, Table 3, and Table 4 ETM as in the broadcast scenario. However, the gain
take only into account the luma component. It can of libaom over the HM is higher than in the
be observed that the results in PSNRY for libaom and broadcast scenario (3.6%), which is not observed
ETM are very close to their PSNRYUV, while VTM for the VTM and ETM. One possible reason is that
PSNRYUV is 3.6% above its PSNRY. This explains why the two-pass encoding can take more benefits from
the differences of ETM and libaom with VTM are a longer intra refresh period, by a better adaptation
smaller for VMAF and MS-SSIM. The superior of the GOP structure, while for HM, VTM and ETM,
performance of VTM chroma tools is not reflected. the GOP structure remains static. It can be also
It should also be noted that only VMAF has a noted that libaom software has a more consistent
temporal dimension. performance over the different picture resolutions
in the streaming senario.
Table 2 – Broadcast: PSNRY BD-rate versus HM
Table 5 – Streaming: PSNRYUV BD-rate versus HM
Streaming
PSNRY VTM8 libaom ETM4 Streaming
UHD −38.7% −18.0% −28.2% PSNRYUV VTM8 libaom ETM4.1
HD −31.7% −15.3% −17.0% UHD −41.2% −20.5% −27.7%
WVGA −28.8% −13.5% −16.7% HD −36.8% −18.5% −18.1%
WQVGA −27.3% −14.1% −16.0% WVGA −31.0% −16.1% −17.2%
Overall −32.4% −15.5% −20.3% WQVGA −29.1% −17.2% −16.4%
Diff vs VTM 0.0% 16.9% 12.1% Overall −35.3% −18.3% −20.6%
Diff vs VTM 0.0% 17.0% 14.7%

80 © International Telecommunication Union, 2020