Page 135 - 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

6.4 L-1 filter 7.1.2 Commercial implementation
The L-1 filter, whose concept is similar to a simple A second set of tests has been performed using a
deblocking filter, can be applied on the L-1 residuals commercial LCEVC implementation provided by
if the transform with the larger kernel size (4x4) is V-Nova, which uses as a base codec x264 for
used. The residuals on the outer boundary of the H.264/AVC and x265 for H.265/HEVC. This set of
transform block are multilplied with a coefficient tests has been performed using a very slow preset
between 0 and 1. The value of these coefficients can for both encoders. The same settings have been
be signalled independently for edges and corners. used when encoding x264/x265 at full resolution as
an anchor and at quarter resolution as a base codec
6.5 Temporal prediction
for LCEVC. The default constant rate factor (CRF)
The temporal prediction uses a zero-motion vector quality setting was used for both x264 and x265,
prediction with a temporal buffer which stores both at full resolution and at a quarter resolution.
residuals from the previous frame only. The 7.2 Objective and subjective metrics results
decision, where to use temporal prediction, is done
on a transform block basis. Additionally, an entire The performance results are shown using the two
tile of 32x32 residuals can be signalled to be used objective metrics Peak Signal-to-Noise Ratio
without temporal prediction, reducing the (PSNR) and Video Multimethod Assessment Fusion
signalling overhead for, e.g., a fast moving part of (VMAF) [8][9]. The latter is operated using the
the sequence. default model (v0.6.1) for HD sequences and the 4K
model (4k_v0.6.1) for UHD sequences.
6.6 Entropy encoding
All the results compare a full resolution video
The two coefficient layers and the temporal layer encoded using the anchor codec (H.264/AVC or
are processed independently by an entropy encoder H.265/HEVC) against a full resolution video
before the encapuslation into the bitstream. The encoded using LCEVC. When comparing against an
entropy coding process consists of two H.264/AVC anchor, the LCEVC would use
components: a Run Length Encoder (RLE) and a H.264/AVC as the base codec. When comparing
Prefix Coding encoder. Aditionally, it is possible to against an H.265/HEVC anchor, the LCEVC would
only use the RLE for the entire data within a use H.265/HEVC as the base codec.
coefficient group.
Two video data sets have been used, Set A for HD
7. PERFORMANCE RESULTS and Set B for UHD resolutions. The sequences of
each data set are listed in Table 1. Each sequence
In order to show that LCEVC is a low-complexity has been tested at four different operating points.
scheme enabling performance improvements over
a given base codec, several tests were run on LCEVC Table 1 – Test sequences
using H.264/AVC and H.265/HEVC as video coding
standards for the base codec. Seq. Sequence name Resolution Frame
rate
7.1 Experimental set-up A1 Campfire 3840x2160 30

The tests have been performed using two different A2 ParkRunning3 3840x2160 50
software implementations. A3 FoodMarket4 3840x2160 60
A4 Fortnite (Part 1) 3840x2160 60
7.1.1 Reference implementation B1 Cactus 1920x1080 50
A first set of tests has been performed using the B2 BasketballDrive 1920x1080 50
LCEVC reference implementation in its current B3 RitualDance 1920x1080 60
version LTM 4.0 [5]. LTM 4.0 uses as a base codec B4 EuroTruck Simulator 1920x1080 60
either reference implementation JM 19.0 for Table 2 provides the average coding performances
H.264/AVC [6] or HM 16.18 for H.265/HEVC [7]. of LCEVC using the Bjøntegaard metric (BD-rate)
[10]. The encodes are performed using the
reference implementation as described in
Section 7.1.1.

© International Telecommunication Union, 2020 113