Duration Dataset¶
Allows you to see the impact of video length on compression and energy consumption.
Name |
Value |
Description (link) |
Total |
|---|---|---|---|
video |
\(100\) slices |
\(100\) |
|
profile |
|
\(2\) |
|
encoder |
|
\(2\) |
|
quality |
\(\left\{\frac{k}{3} \mid k \in [1,2] \right\}\) |
\(2\) |
|
effort |
|
\(2\) |
|
mode |
|
\(2\) |
|
thread |
\(\left\{ 8 \right\}\) |
\(1\) |
|
repeat |
\(10\) times |
total repetitions |
\(10\) |
ramdisk |
|
shm in enc_cmd |
\(1\) |
host |
|
\(2\) |
|
metrics |
|
||
total |
\(\prod\) |
total number of points |
\(27054\) |
Download¶
mendevi download duration.db.xz.torrent
duration.json.xz (24.3 Mo).
duration.db.xz.torrent (29.8 Go).
Plots¶
Encoding energy per frame as a function of the number of frames¶
mendevi plot duration.db -x nbr_frames -y energy_per_frame -e ref_stem -c '(encoder, effort)' -wy profile -wx hostname -t t_enc_encode
Decoding energy per frame as a function of the number of frames¶
mendevi plot duration.db -x nbr_frames -y energy_per_frame -e ref_stem -c '(codec, effort)' -wy profile -wx hostname -t t_dec_decode
Bitrate as a function of the number of frame at constant quality¶
mendevi plot duration.db -x nbr_frames -y rate -c encoder -m effort -wx quality -wy profile -f "mode == 'vbr'"
Conclusion¶
When the video contains enough images, the average energy per frame no longer depends on the duration of the video. Thus, for videos with more than 300 images, ie more than 10 seconds, this effect becomes negligible.
These results are consistent with the UIT-R BT.500-15 (05/2023) which recommends 10-second sequences.
Reproduce¶
First, generate the video segments¶
The first step is to extract random length segments from a video.
The video used is bbb.mp4, the length of the segments follows a base-10 exponential probability distribution.
Next, perform the measurements¶
mendevi encode -r10 -c libx264 -c libsvtav1 -e fast -e medium -m cbr -m vbr -n2 reference*
mendevi probe --no-vmaf reference* sample*
mendevi decode -r10 sample*