IEEE Consumer Electronics Magazine - September/October 2022 - 32

Video Technology
Table 3. Command Lines of Software Encoders.
Encoder
x264
x265
rav1e
Command line
./x264 -demuxer lavf -input-csp yuv420p -profile main -fps 30 -input-res 2560x1600 -psnr -b 0 -I 1
-q 22 -psy-rd 0 -ipratio 1 -no-psy -v -o Traffic_22.h264 Traffic.yuv 2>Traffic_22.txt
ffmpeg -s 2560x1600 -y -r 30 -i Traffic.yuv -c:v libx265 -profile:v main-intra -psnr -x265-params qp
¼ 22:psy-rd ¼ 0:ipratio ¼ 1 Traffic_22.hevc 2>Traffic_22.txt
./rav1e Traffic.y4m -psnr -quantizer 75 -keyint 1 -verbose -o Traffic_22.ivf 2>Traffic_22.txt
Class A: (25001600).
Class B: (19201080).
Class C: (832480).
Class D: (416240).
Class E: (1280...720).
Next, Table 3 shows an example command
Class F: this class is not homogeneous
with regards to spatial resolution (832480,
1024768, and 1280720).
During the video encoding on both embedded
systems, data for consumption and temperature
were gathered.
Setup
Each sequence was encoded using six different
quantization parameter (QP) values to
obtain a wide rate-distortion (RD) range. The QP
values used for H.264/AVC and HEVC were 22,
27, 32, 37, 42, and 47.18
However, the QP values were different for the
AV1 standard implementation proposal (rav1e),
since the range of this parameter is different in
this case, taking values from 0 to 255. To make
a fair analysis between H.264/AVC, HEVC, and
AV1, a QP value approximation was carried out
for the rav1e encoder so that it would cover a
similar range to the rest of the encoders. Thus,
the QP parameter values chosen for the rav1e
encoder in this article were 75, 100, 125, 150,
175, and 200.
In the case of omxh264enc, it may happen that
six encodings per sequence are not performed
because the maximum bitrate of Raspberry hardware
encoding is constrained to 20 Mb/s. Furthermore,
it is not allowed to enter the QP as an input
parameter. Consequently, instead of varying the
QP, the target bitrate was fixed as an input parameter,
using the resulting bitrate from nvv4l2h264enc
for each test sequence.
32
line for each of the three software encoders
used. In this case, Traffic sequence with QP 22
has been used as an example. The same
parameters have been configured in the three
command lines: information related to the
sequence (resolution and framerate), QP, and
intraprediction.
Besides, the three encoders have a command
that can activate acceleration techniques in encoding.
In x264 and x265, the command is " -preset, "
while in rav1e it is " -speed. " These commands
optimize the tradeoff between computing time and
compression efficiency using speed-up encoding
techniques. In the case of x264 and x265, the
" medium " preset has been used, which is the
default mode and does not activate acceleration
techniques in intraprediction. In rav1e, the default
value (6) has alsobeenused.
In conclusion, the comparison is considered
fair since all the command line parameters have
been configured in the same way and intracoding
acceleration techniques have not been activated.
Compression Efficiency Results (Bitrate Versus
PSNR).
The parameters analyzed in this section are
the encoded video bitrate together with its
objective quality, measured with the peak signalto-noise
ratio (PSNR) metric. An encoder is
regarded as more efficient the lower the bitrate
is, and the larger the PSNR, and its performance
is judged using the well-known RD curves. The
curve of the encoder that displays the best performance
will be above that of the others.
Figure 3 shows the performance results for
the Traffic sequence, which belongs to class A. It
can be observed that rav1e presents the best
performance in terms of quality/bitrate, followed
by x265 and nvv4l2h264enc. The encoders that
IEEE Consumer Electronics Magazine

IEEE Consumer Electronics Magazine - September/October 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - September/October 2022