Improve export performance
Brought to you by:
gt67
Menu
▾
▴
#61 Improve export performance
I just did a benchmarking of current svn version (with my patched x264 export fix, see in "patches") versus ffDiaporama devel version (downloaded from here https://forums.linuxmint.com/viewtopic.php?f=47&t=229333&p=1213312&hilit=ffdiaporama#p1213146).
Same project (just 10 photos), 3 seconds each, no audio, crossfade transition, same output format and quality: ffDiaporama takes 26 seconds, imagination 4 minutes...
Maybe something is not working propertly...? It is a huge performance difference for the same task and result.
Related
Imagination is just a graphic frontend to ffmpeg and avconv. Would you please see what happens from the terminal window? Thanks.
On my system using top I see that imagination is running at around 98% while ffmpeg at 20-30% (with 6 or 7 threads working at 5-6%), at around 5-6 fps. So I can conclude that imagination is feeding the pipe to ffmpeg too slowly...
If I try to encode a video with ffmpeg with the same options (x264 -crf 18) it is working at full speed at about 380% cpu (with the same 6 or 7 threads at 40-50% each) at around 35 fps.
Doing a comparison with ffDiaporama is not easy because it is not using ffmpeg as an external process but directly libavcodec etc. However it is encoding at 270% (with 5 threads working) at 30-35 fps, so much more efficiently.
Are you using transitions effects on both softwares?
Yes same project on both, 10 photos with crossfade. See my first post.
How does ffdiaporama achieve the cross fade? Imagination uses cairo that's
notable slow.
On 14 February 2017 at 22:11, Davide Capodaglio davidecapod@users.sf.net
wrote:
Related
Feature Requests: #61
Last edit: Giuseppe Torelli 2018-09-24
Well.. I have no idea! But being a QT project I suppose it uses some Qxxx library, but not cairo.
To sum up, a 10x performance difference is so big that it should be investigated...
For projects that have more than just a few photos, imagination needs... too much patience to wait :-)
Code is here https://svn.tuxfamily.org/viewvc.cgi/ffdiaporama_ffdiaporama/trunk/
Hi David, I just tried to export a MP4 video with your patch and r617 and I didn't notice any significant delay. Would you please do the same ? Thanks.
Hi Giuseppe,
just downloaded and built a snapshot from trunk.
Tried to generate a x264 at 1080p with 10 photos with crossfade, no audio: more than 10 minutes... too slow...
Hi Giuseppe,
just downloaded and built a snapshot from trunk.
Tried to generate a x264 at 1080p with 10 photos with crossfade, no
audio: more than 10 minutes... too slow...
Il 24/09/2018 10:17, Giuseppe Torelli ha scritto:
Related
Feature Requests: #61
Hi Davide,
I used 4 very high res pictures wuth H264 export and 4 transitions, this is the ffmpeg cmd line:
Running /usr/bin/ffmpeg -f image2pipe -vcodec ppm -i pipe: -an -r 25 -y -vcodec libx264 -crf 18 -acodec libmp3lame -ac 2 -ar 44100 -b:a 128k -s 1920x1080 -vf setdar=4/3 "/home/gt/projects/imagination/unknown.mp4"
I can't reproduce the bug, Imagination is very fast in encoding the video. Would you please give me some more info about your system, CPU speed, RAM, ffmpeg version used or avconv?
Thanks
Giuseppe
Last edit: Giuseppe Torelli 2018-09-27
Ciao Giuseppe,
but how long does it take for you to encode 10 big pictures with h264 at
1920x1080 resolution?
My system is a i5 2550k 3.4ghz, 8 gb ram, on Ubuntu 18.04 (but it was the
same with previous 16.04).
I still see the same from my original bug report:
On my system using top I see that imagination is running at around 98%
while ffmpeg at 20-30% (with 6 or 7 threads working at 5-6%), at around 5-6
fps. So I can conclude that imagination is feeding the pipe to ffmpeg too
slowly...
Il giorno gio 27 set 2018 alle ore 20:47 Giuseppe Torelli gt67@users.sourceforge.net ha scritto:
Related
Feature Requests: #61
Ok, I can reproduce the bug. It took even longer than yours, almost 5 minutes. Thanks for this David. Hope to fix it soon, I want to release 3.1 by October.
Hi Davide. Can you update to the current revision, 632, and let me know what you think? There is still work for improvement however. Thanks.
Ciao Giuseppe,
I did a test and I got a strange result.
Same benchmark: 10 photos, 3 seconds each, no audio, crossfade
transition, output at h264 1920x1080 16:9.
Took a total of 2 minutes: the first two photos took 45 seconds each, at
a speed similar to the old version, with low CPU usage with imagination
at 100% and ffmpeg at 20%; then there was a "boost" in speed and the
remaining 8 photos in less than a minute, with ffmpeg at 300% cpu and
imagination at 50%.....
Does it mean anything to you?
Il 08/10/2018 23:40, Giuseppe Torelli ha scritto:
Related
Feature Requests: #61