Page 103 - ITU Journal, ICT Discoveries, Volume 3, No. 1, June 2020 Special issue: The future of video and immersive media
P. 103
ITU Journal: ICT Discoveries, Vol. 3(1), June 2020
Table 6, Table 7, and Table 8, which are based on Globally the subjective quality of the tested
metrics computed only on the luma component, solutions is better than the HM. For VTM and ETM
show the same tendancy as in the broadcast the blocking artefacts visible in the HM on uniform
scenario. The gap of ETM and libaom with VTM is areas with low level variations dissapear. However,
smaller than on Table 5, but in a smaller proportion on static textured areas some experts notice a loss
than in the broadcast scenario on VMAF and of details which is not perceived as a degradation by
MS-SSIM. others. For libaom, a smoother definition and loss of
sharpness is globally observed.
Table 6 – Streaming: PSNRY BD-rate versus HM
Formal subjective tests are required to really
Streaming evaluate the bit rate savings. JVET will conduct a
PSNRY VTM8 libaom ETM4 formal evaluation of VVC with regards to HEVC as
UHD −38.7% −18.0% −28.2% JCTVC did for HEVC with regards to AVC [3]. HEVC
HD −31.7% −15.3% −17.0% was measured at −44% at PSNR BD-Rate, but at
WVGA −28.8% −13.5% −16.7% −59% at MOS BD-Rate (MOS: Measure Of
WQVGA −27.3% −14.1% −16.0% Satisfaction score in subjective tests).
Overall −32.4% −15.5% −20.3% 5.3 Processing time
Diff vs VTM 0.0% 16.9% 12.1% All simulations are run in single thread on the same
platform in order to get comparable encoder and
Table 7 – Streaming: VMAF BD rate versus HM
decoder runtimes.
Streaming The platform characteristics are:
VMAF VTM8 libaom ETM4
UHD −41.6% −21.5% −33.3% Table 9 – Platform characteristics
HD −38.5% −22.6% −24.6% CPU type Intel(R) Xeon(R) Gold 6142
WVGA −29.5% −16.0% −20.2% Hyper threading Off
WQVGA −32.8% −24.0% −23.1% Turbo mode On
Overall −36.4% −21.2% −26.1% Compiler gcc 6.3.0
CentOS7
OS
Diff vs VTM 0.0% 15.2% 10.3% SIMD options SSE42
Table 8 – Streaming: MS-SSIM BD-rate versus HM Table 10 and Table 11 report the runtime factor for
Streaming encoding and decoding versus the HM-16.18 for all
MS-SSIM VTM8 libaom ETM4 sequences. The runtime provides an estimate of the
complexity. “N%” means “N/100” times the
UHD −38.0% −17.0% −28.8% HM-16.18 runtime. The runtimes for the two
HD −30.4% −13.6% −14.8% scenarios, Broadcast and Streaming, are very
WVGA −28.2% −12.1% −16.5% similar.
WQVGA −26.2% −13.6% −16.1% Table 10 – Encoding runtime versus HM
Overall −32.9% −14.5% −20.9% VTM8 libaom ETM4.1
Diff vs VTM 0.0% 18.4% 12.0% Encoding
Broadcast 1308% 497% 669%
5.2 Subjective quality Streaming 1283% 515% 680%
Formal subjective tests have not been conducted to Table 11 – Decoding runtime versus HM
compare these compression solutions, but an Decoding VTM8 libaom ETM4.1
experts’ viewing has been performed to collect
some subjective observations. The broadcast case Broadcast 192% 76% 162%
was taken to compare the HM at a given bit rate per Streaming 201% 77% 167%
sequence with VTM, libaom and ETM at
approximatively the bit rate saving reported in The libaom software has the lowest runtime but as
Table 1 per resolution. The HM bit rate has been mentioned above, the maturity of the software is
chosen at the bit rate point where artefacts may much higher than that of the others. Moreover, they
are not using the same code optimization.
appear.
© International Telecommunication Union, 2020 81