[Ffmpeg-devel-irc] ffmpeg.log.20160812

burek burek021 at gmail.com
Sat Aug 13 03:05:01 EEST 2016


[00:02:09 CEST] <Chloe> can I specify multiple fonts for the drawtext filter? (e.g. a main font and a fallback font)
[00:04:04 CEST] <VamoMenem> Someone know the VCE h264 encoding support status on ffmpeg in linux?
[00:07:06 CEST] <Chloe> VamoMenem: nonexistent afaict
[00:08:32 CEST] <furq> it's not supported on any OS
[00:08:40 CEST] <jkqxz> It kindof works a bit through vaapi with the latest mesa, but it's not properly tested at all.
[00:08:59 CEST] <jkqxz> (Definitely inadvisable to use seriously for anything.)
[00:11:36 CEST] <jkqxz> ((In a "may randomly nuke your machine" type of way, not just "may have dodgy quality".))
[00:21:30 CEST] <VamoMenem> thanks guys. im now choosing if use i7 or A10 cpu for fastest encoding. I think if VCE isnt supported yet i7 is best
[00:22:34 CEST] <VamoMenem> i have 7 ffmpeg live encoders run on an i7 2600k and i need encode 21 simultaneous live streams
[00:25:29 CEST] <furq> what resolution
[00:26:57 CEST] <viric> what are live encoders?
[00:28:04 CEST] <furq> VCE will only run 16 streams anyway, and apparently it's only good for 3 at 1080p30
[00:30:15 CEST] <VamoMenem> im now compressing to 480:270 from dvb-t tv channels in 1080p
[00:30:56 CEST] <VamoMenem> to send it to ms azure cloud
[00:32:11 CEST] <VamoMenem> viric: i call live encoder an a ffmpeg instance recoding from dvb-t to hls in live
[00:33:15 CEST] <viric> ok
[00:33:23 CEST] <viric> you have 7 and you need to reach 21
[00:35:43 CEST] <VamoMenem> yes, im using a i7 2600k but only suports 8 streams top at 75% cpu in total. i use the 8th stream to tandem between others.
[00:36:47 CEST] <VamoMenem> i start the 8th process in the same channel of one of the others, and switch them. To prevent a/v syncs problems.
[00:42:07 CEST] <VamoMenem> i dont know if enabling NVENC and using a nvidia cards can make some diference to avoid buying a lot of computers
[00:42:23 CEST] <furq> nvenc is limited to two concurrent streams unless you buy a quadro
[00:42:44 CEST] <VamoMenem> using a quadro how much streams can handle?
[00:42:56 CEST] <furq> more than two
[00:43:01 CEST] <furq> good luck finding any concrete information about it
[00:43:11 CEST] <VamoMenem> lol
[00:43:23 CEST] <furq> the most solid information i've found about nvenc is angry forum posts
[00:43:50 CEST] <agrathwohl1> VamoMenem I can report anecdotally that, with nvenc, I have been running encodes at 2x the speed I'd get from using libx264.
[00:43:52 CEST] <VamoMenem> yes i try to compile in debbian jessie, two diferent times
[00:44:30 CEST] <agrathwohl1> And I get about half of the CPU usage.
[00:44:48 CEST] <VamoMenem> agrathwohl1: how much encodes you think i can run in parallel?
[00:45:36 CEST] <VamoMenem> i need to encode in 720p from 1080p , 21 simultaneous streams
[00:45:37 CEST] <agrathwohl1> VamoMenem: I haven't tried to push the upper limits. I have done only 2 encodes in parallel under this configuration.
[00:45:45 CEST] <agrathwohl1> But it handled this task more than adequately
[00:46:22 CEST] <furq> VamoMenem: do you mean 720p or 270p
[00:46:24 CEST] <agrathwohl1> My streams were 3840x2160
[00:46:26 CEST] <furq> there's obviously a big difference
[00:46:37 CEST] <VamoMenem> in 720p
[00:47:09 CEST] <VamoMenem> i think it is a huge work
[00:47:49 CEST] <agrathwohl1> VamoMenem I would suppose that with 2 quadros it may be possible to do somewhere near the number of parallel encodes you're after, but I personally feel that 21 is a bit ambitious. Mind you, I am just an experienced FFmpeg user, not a dev.
[00:48:48 CEST] <furq> VamoMenem: i take it you're using -preset ultrafast with x264
[00:49:04 CEST] <VamoMenem> yes, im using that
[00:50:15 CEST] <VamoMenem> im now forcing iframes , then appling a vscale filter and then coding whit libxh264
[00:50:48 CEST] <VamoMenem> and segmenting to hls
[00:52:27 CEST] <VamoMenem> agrathwohl1: which quadro are you using_
[00:54:05 CEST] <agrathwohl1> VamoMenem I am using one Quadro M4000
[00:55:42 CEST] <VamoMenem> a powerfull one
[00:56:02 CEST] <VamoMenem> in linux or windows_
[00:56:50 CEST] <VamoMenem> i can't compile NVENC in debian jessie yet.
[00:58:13 CEST] <agrathwohl1> VamoMenem Linux... I am using the ffmpeg-full-nvenc package from the Arch Linux AUR.
[00:58:43 CEST] <agrathwohl1> nVidia's website has a PDF file which contains pretty easy and thorough instructions on how to compile for Ubuntu.
[01:01:31 CEST] <VamoMenem> agrathwohl1: yes i followed two times the pdf but the patch they suggest don't apply to debian
[01:01:54 CEST] <VamoMenem> im now looking arch linux
[01:02:21 CEST] <VamoMenem> i think i can start using them
[01:02:29 CEST] <furq> you shouldn't need to patch anything afaik
[01:03:30 CEST] <VamoMenem> i dont remember a lot, but in one step when perfrom a git clone of ffmpeg, then you have to path it
[01:03:43 CEST] <furq> that pdf looks very outdated
[01:03:49 CEST] <furq> --enable-nvresize isn't a configure option any more
[01:04:15 CEST] <VamoMenem> yes, is very outdated
[01:04:36 CEST] <VamoMenem> all the versions dont match
[01:05:09 CEST] <furq> https://github.com/FFmpeg/FFmpeg/blob/master/configure#L5776
[01:05:15 CEST] <furq> it looks like all you need is nvencodeapi.h
[01:07:23 CEST] <VamoMenem> maybe i can bought a quadro and test how much streams i can process
[01:08:03 CEST] <agrathwohl1> VamoMenem: The Arch Linux ffmpeg AUR packages are very well maintained, generally. It's a very effective way to get the most out of FFmpeg
[01:08:03 CEST] <VamoMenem> i really in doubt, if buying two i7, or four A10, or two Quadros
[01:08:32 CEST] <VamoMenem> agrathwohl1: you use in arch linux or use it on other distro?
[01:08:49 CEST] <agrathwohl1> VamoMenem I use it in Arch Linux. I have also used it in Gentoo.
[01:09:06 CEST] <agrathwohl1> But setting it up in Gentoo may have been the worst week of my life. ={
[01:09:18 CEST] <VamoMenem> LOL
[01:09:23 CEST] <furq> that sounds like gentoo
[01:09:24 CEST] <VamoMenem> i know that felling
[01:09:30 CEST] <tdr> agrathwohl1, at least it only took a week
[01:09:40 CEST] <agrathwohl1> LOL
[01:10:47 CEST] <VamoMenem> agrathwohl1: you can do a top or something similar for the gpu usage ?
[01:11:37 CEST] <agrathwohl1> VamoMenem There are a few ways to do it. nvidia-settings is a GUI application that ships with the nVidia proprietary drivers. That tool will show you the GPU resource usage percentages and update them in real time (every 5s or so.)
[01:12:14 CEST] <agrathwohl1> Additionally, there is the the tool that the nVidia PDF lists
[01:12:20 CEST] <agrathwohl1> That's a console tool
[01:12:36 CEST] <agrathwohl1> forget the name at the moment
[01:14:14 CEST] <VamoMenem> nice, i will research about that command, i dont want to install x
[01:15:00 CEST] <VamoMenem> the server will be living in a roof
[01:15:28 CEST] <c_14> nvidia-smi
[01:16:24 CEST] <VamoMenem> thanks c_14
[01:16:53 CEST] <VamoMenem> and thanks all you guys!
[02:57:07 CEST] <UnfixedSNIPERJ> hey guys, is there a way to output a converted image as base64 instead of saving the image to a file? I currently am using ffmpeg -i "data:image/gif;base64,b64data" newfile.webm
[02:59:32 CEST] <c_14> ffmpeg -i blah pipe:0 | base64
[03:01:25 CEST] <UnfixedSNIPERJ> so something along the lines of ffmpeg -i animated.gif video.webm pipe:0 | base64
[03:01:42 CEST] <c_14> without the video.webm
[03:01:45 CEST] <c_14> unless you want to output both
[03:01:52 CEST] <c_14> and you'll have to add the format and codec you want
[03:01:58 CEST] <c_14> because ffmpeg can't guess from just pipe
[03:04:31 CEST] <UnfixedSNIPERJ> hmmm I attempted ffmpeg -i animated.gif -c:v libvpx -crf 10 -b:v 1M -c:a libvorbis pipe:0 | base64
[03:04:34 CEST] <UnfixedSNIPERJ> Unable to find a suitable output format for 'pipe:0'
[03:04:34 CEST] <UnfixedSNIPERJ> pipe:0: Invalid argument
[03:04:50 CEST] <c_14> add -f webm
[03:05:38 CEST] <UnfixedSNIPERJ> awesome :D
[03:05:43 CEST] <UnfixedSNIPERJ> thank you :)
[04:50:37 CEST] <UnfixedSNIPERJ> hey guys, so I have: ffmpeg -i "data:image/gif;base64,b64data" -f webm - | base64    This outputs the webm in base64 as expected, is there a way to only show the result base64 and suppress the header/ffmpeg info/data?
[04:52:19 CEST] <c_14> -loglevel quiet
[04:53:21 CEST] <UnfixedSNIPERJ> much appreciated, thank you c_14! :D
[06:27:54 CEST] <robert_> so I'm trying to stream to twitch without audio, because it's over RDP, and I'm wondering how to get anullsrc streaming in AAC format
[06:39:38 CEST] <c_14> -f lavfi -i anullsrc -c:a aac -b:a 16k
[06:40:46 CEST] <robert_> so I can't use it with x11grab?
[06:41:13 CEST] <robert_> and thanks for the response, c_14 :D
[06:44:51 CEST] <c_14> sure you can
[06:45:00 CEST] <c_14> just add that to whatever command you already have in the middle
[06:45:52 CEST] <robert_> oh
[06:46:11 CEST] <robert_> oh, I see
[06:46:34 CEST] <robert_> lafv picks up the anullsrc and x11grab picks up the video
[06:46:35 CEST] <robert_> got it
[06:51:15 CEST] <robert_> c_14: http://sprunge.us/gOaf
[06:51:51 CEST] <c_14> What command are you using?
[06:52:36 CEST] <c_14> Ah, no I see it.
[06:52:46 CEST] <robert_> hm?
[06:52:46 CEST] <c_14> >16 bits per pixel screen is not implemented.
[06:52:54 CEST] <c_14> from x11grab
[06:52:55 CEST] <robert_> hm
[06:52:57 CEST] <c_14> whatever that means
[06:53:08 CEST] <robert_> indeed
[06:53:26 CEST] <robert_> do I need to update something?
[06:53:30 CEST] <c_14> The other problem is that you're copying the video and rawvideo isn't supported in flv.
[06:53:39 CEST] <c_14> First try changing the copy to libx264 or something
[06:54:18 CEST] <robert_> oh, there it goes
[06:54:20 CEST] <robert_> I think
[06:58:51 CEST] <UnfixedSNIPERJ> in regards to https://trac.ffmpeg.org/ticket/1347 does this mean that I cannot send an animated gif that is base64 encoded to to processed and outputted out as base64?
[06:59:21 CEST] <UnfixedSNIPERJ> http://ffmpeg-users.933282.n4.nabble.com/Encoding-a-stream-of-base64-image-strings-into-video-td4671426.html this issue seemed to be brought up but no solution was given
[07:00:34 CEST] <UnfixedSNIPERJ> when I use: ffmpeg -i "data:image/gif;base64,64data" -c:v libvpx -crf 12 -b:v 500K -loglevel quiet -f webm - | base64 it outputs the last frame of the animated gif
[07:51:19 CEST] <jookiyaya> is there big difference between x265 1.5 to  x265 2.0
[08:20:40 CEST] <fa0> I've been reading through 'man', can ffmpeg show the time to complete an encode?
[08:22:48 CEST] <fa0> This is all I'm seeing;
[08:22:50 CEST] <fa0> frame= 1718 fps= 68 q=20.0 size=   34019kB time=00:01:12.10 bitrate=3864.9kbits/s dup=2 drop=0 speed=2.86
[08:23:21 CEST] <fa0> lapsed time, but no time to complete... :(
[08:27:25 CEST] <durandal_1707>  yes, I see that too
[09:06:37 CEST] <viric> fa0: there are some scripts that attempt to do that, wrapping ffmpeg. But ffmpeg alone does not do that
[09:08:04 CEST] <fa0> It's a shame it doesn't seems like a very simple task that should be included into ffmpeg, otherwise, I feel like sitting clueless wondering when a job is going to complete...
[09:08:05 CEST] <fa0> hmm odd
[09:08:40 CEST] <viric> patches welcome, I guess
[09:09:11 CEST] <viric> fa0: maybe cv/progress can help?
[09:11:00 CEST] <viric> progress -c ffmpeg
[09:11:02 CEST] <viric> https://github.com/Xfennec/progress
[09:11:11 CEST] <viric> the poor's man progress solution for linux
[09:11:30 CEST] <fa0> progress -c is already a option in ffmpeg?
[09:12:18 CEST] <viric> no
[09:12:22 CEST] <viric> progress is a progam apart
[09:12:58 CEST] <fa0> oh that github...
[09:13:00 CEST] Action: fa0 looks
[09:13:18 CEST] <viric> your distro may include it
[09:14:13 CEST] <fa0> No progress in Slack I'm aware of...
[09:14:31 CEST] <a-a> is it possible to copy ttf attachment streams into output mp4 files?
[09:14:49 CEST] <viric> that's what slack means.
[09:38:47 CEST] <Spring> is there a way to minimize ffplay playback lag when previewing videos? At larger resolutions it chokes a bit.
[09:40:02 CEST] <fa0> thanks all
[09:40:03 CEST] <Spring> playback on regular video players is fine for example but ffplay can both delay the video displaying (while the audio plays) until it processes it, or upon looping introduce a lag before starting again.
[11:27:35 CEST] <cowai> Does rtmp support mpeg2?
[11:28:28 CEST] <cowai> I need to reliably move mpeg2-ts from across europe and have it play on a raspberry pi.
[11:32:19 CEST] <BtbN> rtmp is flv.
[11:32:26 CEST] <BtbN> mpegts is mpegts.
[11:36:02 CEST] <cowai> flv supports h264, so I just hoped it also somehow could support h262
[11:36:56 CEST] <cowai> I have used multicat and aggregatertp/reorderrtp to send reliable rtp over the internet.
[11:36:58 CEST] <BtbN> well, but you will lose the mpeg-ts obviously.
[11:37:17 CEST] <cowai> What I really want is to keep the video and audio
[11:37:24 CEST] <cowai> I dont care for the actual ts.
[11:37:39 CEST] <cowai> Any tips?
[11:37:43 CEST] <BtbN> Why not just send a plain mpeg-ts stream via tcp?
[11:38:25 CEST] <cowai> Do you know of a good server?
[11:38:31 CEST] <cowai> that I could stream to?
[11:38:48 CEST] <BtbN> netcat, (x)inetd, ...
[11:39:08 CEST] <cowai> Yeah I could just use something like that.
[11:39:20 CEST] <viric> doesn't anybody here use a "non linear video editor" FOSS?
[11:39:37 CEST] <viric> I hate that kdenlive messes up my audio synchronisation
[11:39:49 CEST] <viric> Maybe I should just submit good bug reports about that
[11:41:06 CEST] <durandal_1707> shotcut?
[11:43:28 CEST] <viric> is it nice for audio synch?
[12:10:03 CEST] <durandal_1707> dunno, have yet to try it
[12:21:46 CEST] <viric> ah :)
[12:22:19 CEST] <viric> I tried shotcut and it is really tooooo simple.
[12:51:47 CEST] <Chloe> can I specify multiple fonts for the drawtext filter? (e.g. a main font and a fallback font)
[13:12:31 CEST] <xx_pk_xx> hello. Is there anyone that knows how should I change PTS/DTS/Duration times of a subtitle in TS stream? I am decoding SRT subtitles, and I would like to set those times to DVB subtitles (start time, end time, duration etc.). However, I tried a lot of things, and I'm always getting crazy times after encoding AVSubtitles. Is there anyone who would point me to a right direction with timing in TS files? Thanks
[13:15:56 CEST] <jkqxz> Chloe:  I don't think so.  You could combine your two font files into one instead?
[13:18:30 CEST] <Chloe> jkqxz: yeah, I think that's the only solution, thanks anyways.
[13:28:52 CEST] <durandal_1707> viric: functionality?
[13:54:36 CEST] <viric> durandal_1707: right
[14:27:21 CEST] <Mysoft> hum... any hints on how to do multi thread decoding? (what you have to use?)
[14:30:34 CEST] <Mysoft> (with libav :P)
[14:48:00 CEST] <DHE> Mysoft: set thread_count to something before calling avcodec_open
[14:49:32 CEST] <Mysoft> oh
[14:49:38 CEST] <Mysoft> i was just setting slice_count
[14:49:43 CEST] <Mysoft> and it wasnt working hehe
[15:00:45 CEST] <Mysoft> i see... so setting to 0 make it auto :)
[15:00:53 CEST] <Mysoft> thanks :)
[15:04:28 CEST] <zalan_> hej guys does anyone familiar with the following error?
[15:04:30 CEST] <zalan_> tls packet with unexpected length was received
[17:08:14 CEST] <yang3> is this also the avconv chan ?
[17:10:02 CEST] <c_14> no, that's #libav
[17:10:13 CEST] <yang3> thanks
[18:41:07 CEST] <PlanC> is there any way to calculate or guess the muxing overhead in advance?
[18:42:09 CEST] <__jack__> probably, with the spec and the data-to-be-muxed
[18:42:48 CEST] <PlanC> so as far as I know the muxing overhead contains meta data about the different formats in the file, right?
[18:44:12 CEST] <__jack__> that's the idea, with data on "how to play it", like timestamps etc
[18:44:18 CEST] <PlanC> ah right
[18:44:18 CEST] <__jack__> I'm not expert on that, still
[18:44:40 CEST] <PlanC> because the thing is that I'm currently multiplying with a constant to get the muxing overhead
[18:45:29 CEST] <PlanC> but I'm starting to think that it'd be better to add with a constant instead
[18:46:58 CEST] <__jack__> some container clearly depends on the data inside it (mpegts, for instance)
[18:48:31 CEST] <DHE> lacking the above, I'd suggest using existing content to get an idea what the overhead is and calculating an estimate as needed
[18:49:13 CEST] <PlanC> how can I check what the overhead actually contains?
[18:50:07 CEST] <PlanC> I've noticed that a small file (< 300 kb) has about a 0.1% overhead while a larger file (>20 000 kb) has a 0.001% overhead
[18:50:16 CEST] <PlanC> no idea how to actually view the overhead though
[18:50:31 CEST] <PlanC> if I could see what it was then I might be able to calculate it better
[18:52:04 CEST] <__jack__> 0.1% ? that's not mpegts :) the overhead depends obviously on the container used; you might know it, and you may look after its specs
[18:52:28 CEST] <__jack__> (should)
[18:53:07 CEST] <TD-Linux> PlanC, a linear + offset model is probably reasonable.
[18:53:28 CEST] <kepstin> well, the overhead can be calculated as (stream bitrate * per-packet overhead) + global overhead
[18:53:37 CEST] <kepstin> but what those values are for any given format, I dunno :)
[18:54:14 CEST] <TD-Linux> I'd just get two data points and fit :)
[18:54:38 CEST] <PlanC> TD-Linux: what would the offset be?
[18:55:40 CEST] <kepstin> it's possible there'll be some non-linearities too, e.g. if the format has more or less overhead with different packet sizes, but that effect is probably too small to care about most of the time.
[18:55:50 CEST] <TD-Linux> PlanC, plot the small and big raw and muxed sizes, look where it crosses zero
[18:57:28 CEST] <PlanC> I ran some tests by encoding a few random files
[18:57:43 CEST] <PlanC> I then multiplied the output size with the muxing overhead
[18:57:52 CEST] <PlanC> it's almost always around 0.5 kB
[18:58:19 CEST] <PlanC> I guess I can just add a 1 kB constant to the bitrate
[18:59:23 CEST] <TD-Linux> yeah that's what I'd recommend.
[19:00:01 CEST] <TD-Linux> (to the file size, not bitrate)
[19:02:53 CEST] <PlanC> I'll play around with that now and see if it works
[19:07:42 CEST] <PlanC> that didn't quite work out
[19:07:57 CEST] <PlanC> my calculations for smaller files are 1000 bytes away from the actual size
[19:08:36 CEST] <PlanC> DURATION * (BITRATE * 0.125 * 1000) + 1 = calculated filesize in bytes
[19:09:28 CEST] <TD-Linux> bigger offset, smaller slope? :)
[19:09:51 CEST] <PlanC> ah shoot, no that's not the problem
[19:09:53 CEST] <PlanC> checked the logs
[19:10:04 CEST] <TD-Linux> if you plot this in excel you should be able to figure out something
[19:10:12 CEST] <PlanC> it looks like the bitrate is slightly off
[19:10:27 CEST] <PlanC> I'm telling ffmpeg exactly which bitrate to set it to
[19:10:36 CEST] <PlanC> but it's roughly 0.3 kbit/s more than what I've set it to
[19:25:45 CEST] <PlanC> how can I check whether a file is VBR or CBR with ffprobe/ffmpeg?
[19:48:24 CEST] <Pxl_> whats the best way to escape a hyphen in drawtext? I've tried text='D\'ALERTE', text='D'\''ALERTE' , text='D\\\\\'ALERTE', and text='D\u2019ALERTE'
[19:49:05 CEST] <furq> what about text="D'ALERTE"
[19:49:09 CEST] <Pxl_> single quote*, not a hyphen lol. been a long day
[19:49:36 CEST] <Pxl_> ill try it
[19:51:30 CEST] <jnorthrup> [mxf @ 075be940] track 0: could not find essence container ul, codec not currently supported in container
[19:51:32 CEST] <Pxl_> furq, nope. it doesnt like that since the filter is "drawtext=....."D'ALERTE"...."
[19:51:42 CEST] <jnorthrup> im not sure what a ul is
[19:52:26 CEST] <furq> Pxl_: you might be able to drop those double quotes
[20:43:29 CEST] <Ming_> Wow! There's a lot of people in here
[20:44:52 CEST] <Kris545545> hi, I'm trying to compile ffmpeg 3.1 (for Telegram) on Windows 8.1 (msys2) but ./configure has been running for much more than 1 hour, my config.log is here: http://hastebin.com/raw/asudihokab
[20:46:07 CEST] <PlanC> shouldn't take more than 2 minutes to run
[20:46:23 CEST] <PlanC> try removing the majority of the parameters and see if it finishes
[20:46:49 CEST] <furq> looks like opus is breaking it
[20:47:39 CEST] <Kris545545> furq: and how would I be able to fix it?
[20:48:09 CEST] <furq> try without --enable-libopus
[20:48:19 CEST] <furq> no idea how you'd fix it if you need libopus, i've never used msvc
[20:48:54 CEST] <Kris545545> I /think/ voice messages are encoded using opus, but I'll try without
[20:49:23 CEST] <furq> probably
[20:49:28 CEST] <furq> but you'll need to isolate the error in any case
[21:04:47 CEST] <chocolate-elvis> sorry for the lame question, have a source mov that has 4 mono audio channels as separate tracks, need to map the first 2 to left and right and drop 3, 4. is map_channel the best method?
[21:05:42 CEST] <c_14> https://trac.ffmpeg.org/wiki/AudioChannelManipulation#a2monostereo just don't map the other 2 streams
[21:08:50 CEST] <chocolate-elvis> thanks, been staring at this page for too long and was unsure.
[21:18:19 CEST] <mrelcee> heh  once again, the release crashes on me and builting from git sources is more stable for me.
[22:02:42 CEST] <chocolate-elvis> tried to simplify and do this -c:a:0 -c:a:1 -ac 2, but the results was a Left/Right Stereo, but with just the first audio doubled up
[00:00:00 CEST] --- Sat Aug 13 2016


More information about the Ffmpeg-devel-irc mailing list