[Ffmpeg-devel-irc] ffmpeg.log.20161123

[burek]

[00:00:17 CET] <furq> i can tell you what my favourite tetris piece is though
[00:00:23 CET] <kerio> http://cdn.bulbagarden.net/upload/7/74/722Rowlet.png http://cdn.bulbagarden.net/upload/d/d8/728Popplio.png
[00:00:48 CET] <markvandenborre> :-)
[00:00:57 CET] <kerio> popplio = retarded clown seal
[00:01:04 CET] <kerio> rowlet = cute spherical owl
[00:01:04 CET] <furq> i did a google image search for "l block" and the first match is a j block
[00:01:09 CET] <furq> fucking sort it out google
[00:01:40 CET] <kerio> hey furq
[00:01:45 CET] <kerio> i have another question for you
[00:02:01 CET] <furq> what is it my nephew
[00:02:08 CET] <kerio> has anyone really been far even as decided to use even go want to do look more like?
[00:02:12 CET] <furq> yes
[00:02:31 CET] <furq> george stockliffe did it in 1907
[00:05:56 CET] <markvandenborre> somehow, I feel this is not the time for more questions from me :-)
[00:06:10 CET] <markvandenborre> thanks a lot, all, and have a good night!
[00:06:34 CET] <markvandenborre> (or whatever time of day it is at your place)
[00:57:29 CET] <markvandenborre> ok, couldn't sleep, so here I am again
[00:58:02 CET] <markvandenborre> mediainfo shows relevant info about an flv file that ffmpeg says "protocol not found" about
[00:59:14 CET] <saintgermain> On Debian Stable with ffmpeg 3.2.2, I tried the command "ffmpeg -hwaccel vaapi -i castle.mp4 -f null -" and it works without problem. But with ffmpeg static (3.2 or git) I have a bunch of errors "No VA display found for device: ." and "Failed to create a VAAPI device". Did I use hwaccel incorrectly or is it a regression ?
[00:59:20 CET] <markvandenborre> I did the same with this flv as I did before with an mp4: kill -9 the process that feeds the nginx-rtmp server
[00:59:54 CET] <markvandenborre> is there a way to force ffmpeg to use a specific protocol to handle this file?
[01:00:08 CET] <markvandenborre> mpv plays it back without a glitch
[01:00:51 CET] <markvandenborre> vlc too
[01:01:12 CET] <markvandenborre> the ff* suite doesn't find protocol info
[01:02:25 CET] <markvandenborre> obviously, enough information is present in the file to play it back to the same extent as other players than ffplay
[01:03:55 CET] <markvandenborre> ok, that was a dumb question of mine
[01:10:37 CET] <FlyingIT> hello community! is anyone around who has set up a raspberry pi 3 to stream to twitch with ffmpeg? I've been trying to get it work for a few days now, closest I got was to get it to stream with speed=0.5x, so not quite...
[01:18:37 CET] <markvandenborre> FlyingIT: https://datarhei.github.io/restreamer/
[01:18:46 CET] <markvandenborre> might help?
[01:18:53 CET] <markvandenborre> as inspiration I mean
[01:20:59 CET] <saintgermain> On Debian Stable with ffmpeg 3.2.2, I tried the command "ffmpeg -hwaccel vaapi -i castle.mp4 -f null -" and it works without problem. But with ffmpeg static (3.2 or git) I have a bunch of errors "No VA display found for device: ." and "Failed to create a VAAPI device". Did I use hwaccel incorrectly or is it a regression ?
[01:25:02 CET] <jkqxz> saintgermain:  libva does not work when statically linked.
[01:25:38 CET] <saintgermain> jkqxz: so I cannot use the ffmpeg static and I have to compile myself ?
[01:29:32 CET] <jkqxz> You must be dynamically linked to libva.  That doesn't necessarily mean you have to compile it yourself on the target system (after all, the Debian version does work), but in practice it may be the only way of using a specific version.
[01:31:07 CET] <FlyingIT> hey markvandenborre thanks for the link!! You know what, I just changed the fps to half of before and then the speed issue got resolved. I can't believe it was such a simple fix. Wow, I really could have thought about that earlier...
[01:31:33 CET] <saintgermain> jkqxz: Ok thanks !
[01:31:56 CET] <FlyingIT> But often even just asking helps find the solution :D
[01:31:56 CET] <furq> FlyingIT: what command are you using
[01:36:03 CET] <FlyingIT> oh wow, the datarhei restreamer looks cool running on docker!
[01:36:24 CET] <FlyingIT> furq hold on; I'll write it, 1mo
[01:40:53 CET] <markvandenborre> this is strange...
[01:41:58 CET] <FlyingIT> the basic one I got to work now is:
[01:41:59 CET] <FlyingIT> ffmpeg -thread_queue_size 1024 -f v4l2 -r 15 /dev/video0
[01:42:47 CET] <markvandenborre> my debian stable server recognises the .flv file that my debian testing desktop complains about with "protocol not found"
[01:43:05 CET] <markvandenborre> both are using _very_ similar versions of ffmpeg it seems
[01:43:22 CET] <markvandenborre> (my server a backported package from testing)
[01:44:33 CET] <markvandenborre> 7:3.2-2 on testing versus 7:3.2-2~bpo8+2 for stable with backports
[01:44:52 CET] <markvandenborre> (~bpo8+2 meaning a backported version)
[01:50:09 CET] <furq> FlyingIT: that doesn't look like a valid command
[01:50:20 CET] <furq> markvandenborre: those are identical versions aren't they
[01:51:13 CET] <FlyingIT> well I'm running it from a bash file
[01:51:30 CET] <FlyingIT> also I'm running Arch Linux, if that makes any difference, wasn't too happy with raspberian
[01:51:48 CET] <furq> put echo in front of the command in the script
[01:51:59 CET] <FlyingIT> what would this do?
[01:52:07 CET] <FlyingIT> *output?
[01:52:07 CET] <furq> it'll print the actual command
[01:52:12 CET] <FlyingIT> oh, I see
[01:52:40 CET] <markvandenborre> furq: looks like it, yes
[01:52:52 CET] <furq> what's the filename
[01:53:20 CET] <markvandenborre> pingu-23-Nov-16-00:42:19.flv
[01:53:44 CET] <furq> i remember
[01:53:47 CET] <furq> dsf
[01:54:00 CET] <furq> there's some mailing list issue about that error showing up if you have colons in the filename
[01:54:21 CET] <furq> apparently it goes away if you prefix the filename with "file:"
[01:54:30 CET] <FlyingIT> oh yeah! I had the whole output left out of the command, soz. -vcodec libx264 -pix_fmt yuv420p -acodec libmp3lame -threads 2 -preset ultrafast -f flv -ar 22050 rtmp://$INGEST.twitch.tv/app/$STREAM_KEY"
[01:54:40 CET] <furq> yeah don't use x264 on an rpi
[01:54:45 CET] <FlyingIT> no?
[01:54:52 CET] <furq> there's an onboard hardware h264 encoder which sucks, but it should at least run fast
[01:54:53 CET] <markvandenborre> furq: yes, it doesn't
[01:55:07 CET] <furq> -c:v h264_omx
[01:55:11 CET] <markvandenborre> fall over anymore
[01:55:13 CET] <furq> you'll need to compile ffmpeg with it though
[01:57:31 CET] <FlyingIT> oh, so h264_omg is preferable, I'll try now. Also I don't quite understand the threads parameter. 2 seem fine, but I'd have thought 4 would be better?
[01:58:08 CET] <furq> well on a pi 3 you probably want 6
[01:58:15 CET] <furq> with libx264, anyway
[01:58:20 CET] <furq> that option makes no sense with omx
[01:58:56 CET] <markvandenborre> furq: you potentially saved us several hours of fiddling
[01:59:52 CET] <markvandenborre> and you helped us build something more reliable and _simpler_ for FOSDEM. Thanks a lot!
[02:07:02 CET] <FlyingIT> if I use h264_omx I get this warning:
[02:07:03 CET] <FlyingIT> Codec AVOption preset (Set the encoding preset (cf. x264 --fullhelp)) specified for output file #0 (/storage/.kodi/userdata/addon_data/service.emby/transcoding-temp/44802c6446da28c773b15182d74143b1.m3u8) has not been used for any stream. The most likely reason is either wrong type (e.g. a video option with no video streams) or that it is a private option of some encoder which was not actually used for any stream.
[02:07:33 CET] <FlyingIT> The stream works fine though, but that was the reason why I switched back to libx264
[02:08:51 CET] <voip_> hello guys
[02:09:20 CET] <voip_> i have problem on ubuntu : bind failed: Permission denied
[02:09:54 CET] <voip_> when i trying to execute witout sudo
[02:10:36 CET] <furq> you need to be root to bind a port below 1024
[02:11:20 CET] <furq> or you could give ffmpeg CAP_NET_BIND_SERVICE
[02:16:52 CET] <FlyingIT> WOW, when I use libx264 the bitrate=~2000kbits/s and with h264_omx it's ~200kbits/s. Does this mean libx has 10x higher upstream?
[02:17:03 CET] <voip_> furq, thank you
[02:17:09 CET] <furq> no it means they have different defaults
[02:17:22 CET] <FlyingIT> oh! hehe, I see
[02:17:24 CET] <furq> you need to set -b:v to set the bitrate with omx
[02:17:37 CET] <FlyingIT> okay :)
[02:17:39 CET] <furq> it'll probably look worse than x264 ultrafast, but it won't touch your cpu
[02:17:53 CET] <FlyingIT> omx will look worse?
[02:17:57 CET] <furq> yeah
[02:18:09 CET] <FlyingIT> but it'll be more stable?
[02:18:38 CET] <furq> yeah it's a hardware block on the rpi gpu
[02:19:29 CET] <saintgermain> jkqxz: I tried compiling FFmpeg master and "-hwaccel vaapi" works with a reduce CPU load. However I tried compiling FFmpeg branch 3.2, and I got the errors "No VA display found for device: ." and "Failed to create a VAAPI device". In addition "-hwaccel vaapi" works on Debian FFmpeg 3.2.2, but the CPU load is really heavy. Something strange is going on. Any idea ?
[02:19:31 CET] <FlyingIT> okay nice! And does the one warning I get when using omx mean anything?
[02:19:37 CET] <furq> what warning is that
[02:19:49 CET] <FlyingIT> Codec AVOption preset (Set the encoding preset (cf. x264 --fullhelp)) specified for output file #0 (/storage/.kodi/userdata/addon_data/service.emby/transcoding-temp/44802c6446da28c773b15182d74143b1.m3u8) has not been used for any stream. The most likely reason is either wrong type (e.g. a video option with no video streams) or that it is a private option of some encoder which was not actually used for any stream.
[02:19:57 CET] <furq> oh
[02:20:01 CET] <FlyingIT> forget the kodi part
[02:20:02 CET] <furq> you're specifying an option which doesn't do anything
[02:20:20 CET] <FlyingIT> I copied that from somewhere, but the error is pretty much the same
[02:20:29 CET] <furq> probably -preset ultrafast
[02:20:34 CET] <furq> just get rid of that
[02:20:43 CET] <furq> oh yeah it says it's preset
[02:20:45 CET] <FlyingIT> oh, yeah. That's only needed for libx264
[02:21:05 CET] <furq> also you probably want to use aac for streaming over rtmp
[02:21:15 CET] <FlyingIT> oh yeah xD. I thought it meant something with preset... Getting late here I suppose :)
[02:21:25 CET] <furq> and there's no reason to use 22khz unless you're starved for bandwidth
[02:21:43 CET] <FlyingIT> for the audio you mean?
[02:21:55 CET] <furq> y
[02:21:56 CET] <FlyingIT> and I'd have to check if twitch takes aac
[02:22:09 CET] <furq> it should
[02:22:14 CET] <FlyingIT> just a mo
[02:26:04 CET] <FlyingIT> so instead of '-acodec libmp3lame' it's '-acodec aac'. Not complaining and streaming fine :)
[02:26:22 CET] <FlyingIT> furq do you stream stuff?
[02:31:24 CET] <furq> not regularly
[02:32:42 CET] <FlyingIT> same here
[02:37:34 CET] <markvandenborre> furq: wasn't libfdk_aac considerably faster and better quality too?
[02:37:48 CET] <markvandenborre> (although that might require a recompile of the source package with the option enabled)
[02:38:00 CET] <markvandenborre> FlyingIT: might be interesting to you
[02:38:04 CET] <furq> idk about faster but it is higher quality
[02:38:10 CET] <furq> but yeah it's nonfree so it needs a rebuild
[02:38:40 CET] <FlyingIT> does nonfree mean you need a licence or you need to rebuild your ffmpeg?
[02:38:50 CET] <furq> it means you can't distribute binaries with fdk linked in
[02:39:03 CET] <furq> you technically need a license for a lot of ffmpeg encoders but nobody cares unless you're making money from it
[02:39:25 CET] <FlyingIT> hehe, okay. Makes sense! ^^)
[02:45:02 CET] <shout> Greetz, is there an api or best practices available for a non CLI way to define ffmpeg's inputs? Looking for source or docs. Thanks in advance.
[02:46:49 CET] <c_14> https://ffmpeg.org/doxygen/trunk/index.html https://ffmpeg.org/doxygen/trunk/examples.html
[02:50:40 CET] <saintgermain> Debian ffmpeg (3.2) and ffmpeg github checkout "master" have working vaapi hwaccel (but Debian ffmpeg is very slow). Static ffmpeg "git", static ffmpeg "3..2" and ffmpeg github checkout "3
[02:51:18 CET] <saintgermain> Debian ffmpeg (3.2) and ffmpeg github checkout "master" have working vaapi hwaccel (but Debian ffmpeg is very slow). Static ffmpeg "git", static ffmpeg "3..2" and ffmpeg github checkout "3.2" gives error "No VA display found for device: .". Has somebody an explanation ?
[02:55:07 CET] <c_14> Didn't somebody just tell you that libva doesn't work when statically linked?
[02:56:24 CET] <saintgermain> c_14: indeed but I tried compiling FFmpeg 3.2 myself and it doesn't work either. However compiling FFmpeg master works. So there is something else going on.
[02:56:59 CET] <c_14> Then open a bugreport?
[02:57:18 CET] <c_14> Sounds like something was fixed and not backported
[02:57:56 CET] <saintgermain> c_14: But FFmpeg 3.2 from Debian works (albeit slow). That's what is puzzling me.
[02:58:19 CET] <c_14> same build configuration?
[02:58:22 CET] <c_14> same subversion?
[02:58:27 CET] <c_14> does debian patch their release?
[02:59:13 CET] <saintgermain> c_14: good idea, I don't know, I'll check !
[03:01:25 CET] <saintgermain> c_14: nothing in the changelog about vaapi except perhaps "Build static libraries without -fPIC as required by policy 10.2"
[03:06:08 CET] <FlyingIT> thanks furq for your help! :) I'm almost off too sleep now, cya around o/
[03:06:23 CET] <FlyingIT> have a nice night everyone
[03:27:49 CET] <FlyingIT> okay, tried out one more thing. 1 camera running for stream works perfectly, but 2 cameras seem to be too much for the raspberry pi 3... Maybe I'll find a way around this to tweak it, but not tonight xD.
[03:28:12 CET] <FlyingIT> Maybe I can set -threads?
[03:28:29 CET] <FlyingIT> Anyways, safe! I'm off now for sure
[04:12:17 CET] <Sircle> Can anyone help me with this? https://pastebin.mozilla.org/8930716
[07:44:53 CET] <LongChair1> Hello :)
[07:45:34 CET] <LongChair1> i found tracks of some v4L2 m2m work for ffmpeg in mailing lists https://ffmpeg.org/pipermail/ffmpeg-devel/2014-November/165630.html
[07:45:53 CET] <LongChair1> doesn anyone knows if this work is still going on ?
[08:37:20 CET] <Sircle> Can anyone tell how to get all frames via this snippet? This is only getting the first frame http://stackoverflow.com/a/15736494
[09:52:41 CET] <alexvf> hi, is this the right place to ask about ffmpeg libraries usage?
[09:52:59 CET] <enyc> alexvf: "Welcome to the FFmpeg USER support channel"
[09:53:09 CET] <alexvf> :)
[09:53:22 CET] <enyc> alexvf: I don't see why not, but clearly define your question and be patient...  I probably can't help but somebody may beable to eventually.
[09:53:40 CET] <alexvf> enyc: ok, thank you
[09:54:25 CET] <alexvf> i'm trying to write a program that outputs hls preview images
[09:54:43 CET] <alexvf> i know i can do it with ffmpeg command directly
[09:55:00 CET] <alexvf> but i'm trying to do it more efficiently
[09:55:23 CET] <alexvf> basically avoid having to initialize all for each fragment
[09:56:10 CET] <alexvf> so i start the program with a fragment, which is detected by avformat_open_input and avformat_find_stream_info
[09:56:43 CET] <alexvf> and then i pass other fragments to the program, hopefully not needing another avformat_find_stream_info
[09:57:42 CET] <alexvf> but it crash when i try to reuse the AVCodecContext from the previous fragment
[09:58:12 CET] <alexvf> is what i want possible at all?
[10:00:17 CET] <markvandenborre> alexvf: I have no idea about exactly what you are asking, but...
[10:00:30 CET] <markvandenborre> did you have a look at the nginx-rtmp source code for inspiration?
[10:01:10 CET] <markvandenborre> it has exactly the functionality you are asking for (at least from an flv container)
[10:01:15 CET] <alexvf> no, i started with the famous dranger tutorial, part 1, which outputs a .ppm image and extend it
[10:01:39 CET] <alexvf> idoes nginx-rtmp take hls previews?
[10:02:04 CET] <markvandenborre> have a look at it, I can't help you any further
[10:03:21 CET] <alexvf> ok, thank you
[10:03:39 CET] <alexvf> what i want is very easy, but i'm not good explaining myself
[10:03:41 CET] <furq> nginx-rtmp isn't based on ffmpeg
[10:03:50 CET] <furq> if that matters to you
[10:04:07 CET] <alexvf> i have a HLS stream (several fragments) and i want a preview image for each of them
[10:04:24 CET] <alexvf> but i want to save preview in a efficient way
[10:04:46 CET] <alexvf> if i do it through ffmpeg command it takes more than a second for each one
[10:05:12 CET] <alexvf> and i think it could take much less, since i don't event need to download all the fragment, just to the point of decoding a frame
[10:06:09 CET] <alexvf> furq: do you mean it does not use ffmpeg?
[10:06:27 CET] <furq> it doesn't use the ffmpeg libs for hls
[10:06:32 CET] <furq> i also don't think it has an hls demuxer
[10:06:43 CET] <furq> it has a muxer but it only takes rtmp input
[10:07:13 CET] <alexvf> furq: main problem is that i want to use ffmpeg libraries, that's why i am here :)
[10:18:09 CET] <haspden> Hi, whilst installing FFmpeg, I get issues when running the OPUS installer... I took the install code from https://trac.ffmpeg.org/wiki/CompilationGuide/Centos however get the error on the # make line
[10:18:33 CET] <haspden> tests/test_opus_decode.c:139: error: #pragma GCC diagnostic not allowed inside functions
[10:18:51 CET] <haspden> tests/test_opus_decode.c:140: error: #pragma GCC diagnostic not allowed inside functions
[10:18:56 CET] <haspden> tests/test_opus_decode.c:144: error: #pragma GCC diagnostic not allowed inside functions
[10:19:08 CET] <haspden> make[2]: *** [tests/test_opus_decode.o] Error 1
[10:19:32 CET] <tdr> youre building all those from source on centos?
[10:19:47 CET] <tdr> (all the packages on that page i mean)
[10:19:57 CET] <haspden> yea
[10:20:07 CET] <tdr> must like pain
[10:20:17 CET] <tdr> your centos version may matter, gcc versions change
[10:21:18 CET] <haspden> all I want it ffmpeg to run haha but it's taken me all night here ha
[10:21:21 CET] <haspden> pain indeed
[10:21:39 CET] <haspden> running centos 6
[10:22:04 CET] <tdr> i get it, when useing old old RH i used to rebuild all my media stuff to unbreak it, but it always sucked
[10:22:48 CET] <haspden> 6.8 sorry
[10:22:48 CET] <spaam> 6!? that is like 5 years old
[10:24:23 CET] <haspden> its so frustrating... I want to do something so simple, but have it run server side... seems ffmpeg would be the only way to go however
[10:24:41 CET] <haspden> I want to take a jpg and turn it into a video file
[10:24:48 CET] <haspden> preferably .mp4
[10:25:07 CET] <haspden> a night of trying to install ffmpeg and i'm still no futher forward haha
[10:25:16 CET] <furq> haspden: http://johnvansickle.com/ffmpeg/
[10:44:36 CET] <haspden> well thats far easier
[11:57:41 CET] <alexvf> is there a way of setting up an AVCodecContext without calling avformat_find_stream_info?
[13:20:30 CET] <JonG> This line in FFmpeg's configure help: https://github.com/FFmpeg/FFmpeg/blob/master/configure#L139 ... Does that mean that by default network support is on or off? I am a bit confused by --disable- paired with [no]. How should I read the output to figure out the defaults?
[13:21:24 CET] <BtbN> It's not disabled by default.
[13:24:52 CET] <dragmore88> hi. Anyone that could suggest a syntax taking 2 mono channels out from a prores file and encode them to AAC 2 channel ?
[13:25:06 CET] <dragmore88> this one fails:
[13:25:10 CET] <dragmore88> ffmpeg -i Everest2015_Theatrical_5min_TestClip_HDR_UHD_PRORES.mov -filter_complex "[6:a][7:a]amerge=inputs=2[aout]" -map "[aout]" output.wav
[13:25:20 CET] <dragmore88> that was for pcm
[13:25:33 CET] <dragmore88> Invalid file index 6 in filtergraph description [6:a][7:a]amerge=inputs=2[aout].
[13:27:14 CET] <JonG> BtnB: Thanks. So what is the significance of "[no]". I.e. how does "--disable-network        disable network support [no]" differ from "--disable-dct            disable DCT code"?
[13:27:25 CET] <JonG> BtbN sorry
[13:28:19 CET] <BtbN> from looking at the configure scripts, "network" is just a dummy target that all network protocols like tcp, http, sftp and so on depend on.
[13:28:33 CET] <BtbN> So if you want to get rid of all of them at once, you just disable network.
[14:26:29 CET] <Guest34390> Hi friends, someone already have problems to encode multiple bitrates using ffmpeg and blackmagic with ubuntu, and using CentOS 7 its works? Same compilation, same hardware, but different distros
[14:27:24 CET] <Guest34390> had*
[14:28:24 CET] <DHE> no errors displayed?
[14:29:04 CET] <DHE> if it's the same binary, you may have problems with shared libraries. maybe get a static build of ffmpeg, or strip the features down to just basic things like glibc and zlib
[14:29:13 CET] <Guest34390> yes, the error is: [decklink @ 0x2a187c0] > Decklink input buffer overrun!39kB time=00:00:07.82 bitrate=  41.2kbits/s speed=0.537x
[14:30:28 CET] <DHE> at a speed of 53.7% realtime, yeah you're going to have a bad time with realtime processing
[14:30:42 CET] <Guest34390> yep :'(
[14:31:49 CET] <Guest34390> but with centos its ok! I tryiing to see if there is some tuning that is built in in centOS 7 and isnt in ubuntu... but i not really sure if the main problem is this, I really appreciate suggestions!
[14:39:28 CET] <BtbN> Your CPU is too slow for the encode and filter settings you choose.
[14:39:37 CET] <BtbN> Either get a faster machine, or use faster settings.
[14:40:58 CET] <DHE> or something like a GPU to do accelerated encodnig?
[14:46:26 CET] <Guest34390> The hardware in both machines are identical!
[14:46:45 CET] <Guest34390> but with centos 7 its works great, but with ubuntu, the problem persist!
[15:10:16 CET] <DHE> CPU busy with other things perhaps?
[15:10:43 CET] <Guest34390> nope DHE , this machine is only for this usage!
[15:11:05 CET] <Mysoft> if i set max bitrate limit...... and set maximum crf, it will reduce framerate (skip frames) to try to achieve the desired bitrate?
[15:11:42 CET] <JEEB> depends on the encoder how it deals with "cannot spend bits"
[15:11:47 CET] <JEEB> x264 never removes pictures
[15:12:02 CET] <JEEB> theora if I recall correctly was infamous for dropping pictures
[15:12:21 CET] <Mysoft> i see... that's kinda of a problem
[15:12:46 CET] <DHE> hmm.. seems unlikely for a kernel difference to affect CPU usage so dramatically... which might suggest a hwaccel difference between systems?
[15:12:48 CET] <Mysoft> i would prefer a framerate drop in cases the encoding is too hard (too much motion) instead of a drop in visual
[15:14:35 CET] <Guest34390> we are using libx264 pra vídeo
[15:14:52 CET] <Guest34390> and  libfdk_aac for audio
[15:15:26 CET] <BtbN> there is no hwaccel in decoding and encoding with ffmpeg.c unless it's explicitly requested
[15:15:54 CET] <DHE> well, that's software encoding. x264 does have some opencl support, but the effect isn't strong enough to double the speed
[15:16:21 CET] <BtbN> isn't it like 3% faster, at best?
[15:16:33 CET] <DHE> it does vary. I've seen +20%, I've also seen -10%
[15:16:48 CET] <DHE> so, yeah...
[15:16:50 CET] <Mysoft> depends on how much the lookahead is necessary :)
[15:17:40 CET] <Mysoft> but i think hw accs should be doing more... (not that i would benefit me... since other than cuda/opencl) no other hwacc works here
[15:18:15 CET] <Mysoft> unless someone make a shader based x264 encoder :D
[15:19:58 CET] <Guest34390> I'm really trying to think how to solve or just tunning ubuntu to run like centOS7 but still haven't worked!
[15:20:04 CET] <Guest34390> work*
[16:47:21 CET] <persina> what does "-vf scale=7680:4320,setdar=22:9 " mean?
[16:50:55 CET] <durandal_1707> rescale and change display aspect ratio
[16:52:37 CET] <persina> durandal_1707: If I want to change aspect ratio but not rescale...what would the argument be?
[16:56:36 CET] <durandal_1707> just sample aspect ratio?
[16:58:22 CET] <JonG> persina: Does this help? https://trac.ffmpeg.org/wiki/Scaling%20(resizing)%20with%20ffmpeg
[16:58:34 CET] <skcustibrc> can someone help me understand how to decimate correctly
[16:59:26 CET] <skcustibrc> i have a video at 59.940 fps, but it should be 23.940 to have each rame unique
[17:16:38 CET] <JEEB> hmm, /21
[17:22:28 CET] <pgorley> hi all, which function do i need to use to apply a filter on a frame?
[17:22:39 CET] <pgorley> i'm looking at ffmpeg.c and i can't seem to figure it out
[17:23:30 CET] <pgorley> specifically, i'd like to apply format=nv12 and hwupload in that order
[17:31:21 CET] <jkqxz> hwupload is quite trivial.  Unless you have other filters you want to apply at the same time, it may be simpler to make the hwframe context and transfer directly.
[17:36:21 CET] <pgorley> alright, thanks
[17:43:11 CET] <kerio> would mpeg2 be preferrable to mjpeg?
[17:50:25 CET] <kepstin> kerio: for video? yeah, it supports P/B frames so the result should be quite a bit smaller at the same quality. Obviously it would be better to use a newer coded if possible tho.
[17:50:49 CET] Action: kepstin suspects it would be almost the same as mjpeg if used intra-only?
[17:51:02 CET] <kerio> kepstin: the issue is encoding cpu time
[17:51:15 CET] <kerio> the 3ds is really bad
[17:51:36 CET] <kerio> and the wifi chip it uses is also really bad, so you can't go with raw video either
[17:52:03 CET] <kepstin> maybe you should look at, i dunno, ffvhuff or something like that?
[17:52:33 CET] <kerio> would that really save much, tho
[17:53:04 CET] <kepstin> it's really fast, and depending what the video content is it might save a good chunk...
[17:53:08 CET] <jkqxz> MJPEG is pretty standard as the easiest thing which still gives enough compression to be usable over a slightly-limited channel.  Hence use by every webcam ever, for example.
[17:54:17 CET] <Mavrik> Then again, these days most stuff uses H.264 since every SoC and it's mom has a HW encoder
[17:57:21 CET] <kerio> sadly, the 3ds has nothing of that sort, and the n3ds only has a h264 decoder
[17:57:28 CET] <kerio> otherwise it would've been my first choice
[18:03:38 CET] <kepstin> but yeah, i'd expect mpeg2 to be a bit slower than mjpeg since it's doing more work per frame to detect motion, etc - on the other hand, depending on video content, that work can save encoding work later, so... :/
[18:03:59 CET] <kepstin> have to test and see, really
[18:04:26 CET] Action: kepstin expects that neither the jpeg or mpeg2 encoders in ffmpeg are well optimized for arm
[18:04:50 CET] <kerio> right now this weird thing is using libjpeg i believe
[18:05:01 CET] <persina> I have 2 PNG's with 3840x2880 when i try to make a video from them all i get is a black screen...command and output: http://pastebin.com/7XMgBHCQ
[18:05:21 CET] <kepstin> kerio: which libjpeg? you probably want to use libjpeg-turbo if possible
[18:07:25 CET] <durandal_1707> persina: ffplay plays last two images just fine?
[18:08:42 CET] <persina> durandal_1707: Yes.
[18:08:56 CET] <persina> durandal_1707: Does my media player have a resolution limit?
[18:13:17 CET] <durandal_1707> persina: dunno
[19:41:32 CET] <pgorley> how di convert from one pixel format to another using the libav* libs?
[20:11:43 CET] <dmfrey> Hi all. Not sure if this is the right channel, but is it possible to use one of the libraries, like libavcodec, through JNI on android to decode mpeg2?
[20:12:15 CET] <dmfrey> i can build it and am already using it to play back ac3 audio
[20:13:37 CET] <dmfrey> I plan on writing an extion to ExoPlayer (https://github.com/google/ExoPlayer) that will allow my app to provide native mpeg2 ts playback from within my app
[21:42:03 CET] <Guest78> hey guys . i have a strange question that i need help with. We're trying to stream live from a linux server. The flow that we use is: start Xvfb, start pulseaudio, start chrome, start ffmpeg to flv to youtube. Now, all is cool, but, when ffmpeg starts exactly after the chrome loaded, there is a 3s delay between the audio and video into the youtube video. And now the strange part: restarting FFMPEG or delaying it's start with let's say
[21:42:03 CET] <Guest78>  10s (kind of a buffer for chrome to really fully start) -> 0s delay between audio and video. Any ... ideas ?
[23:11:51 CET] <Guest78> http://stackoverflow.com/questions/40774967/ffmpeg-stream-chrome-kiosk-mode-ubuntu-16-04-server
[23:22:58 CET] <Jan-> hihi all
[23:23:47 CET] <Jan-> I just had someone from a big video equipment company tell me that Quicktime doesn't define a way to set an aspect ratio.
[23:23:59 CET] <Jan-> Surely this is drivel
[23:24:04 CET] <Jan-> you can totally do it with -aspect foo
[23:24:08 CET] <Jan-> and that will be respected by players
[23:26:42 CET] <furq> isn't that set in the h264 sps
[23:27:29 CET] <JEEB> there's multiple things
[23:28:00 CET] <JEEB> video stream -> PASP -> width/height in whatever video info thing there was in ISOBMFF
[23:28:09 CET] <JEEB> and if it's about QTFF then you look at the qtff.pdf
[23:28:18 CET] <JEEB> (QTFF being what we usually call MOV)
[23:28:45 CET] <Jan-> I don't know what the -aspect option sets, to be honest.
[23:28:59 CET] <Jan-> I know that when we set -aspect on *.mov it is respected by VLC, Quicktime and Adobe Premiere.
[23:29:04 CET] <furq> whatever -aspect does doesn't show up in ffprobe -show_format
[23:29:09 CET] <Jan-> I have no idea.
[23:29:12 CET] <JEEB> just saying that there's multiple places aspect ratio can be set in ISOBMFF or MOV
[23:29:19 CET] <JEEB> well, MOV I don't know of :P
[23:29:24 CET] <JEEB> ISOBMFF I've looked through
[23:29:29 CET] <Jan-> I do know that sometimes there's a SAR set in the codec data.
[23:29:35 CET] <Jan-> which can be confusing.
[23:29:36 CET] <JEEB> yes
[23:29:37 CET] <JEEB> https://social.technet.microsoft.com/Forums/windows/en-US/fed910ba-0b25-48f1-8abb-39bc56095831/a-small-bug-in-windows-media-player-12-on-windows-81-and-windows-7?forum=w8itprogeneral
[23:29:48 CET] <JEEB> I actually wrote down the ways to store aspect ratio there
[23:29:55 CET] <JEEB> in ISOBMFF (aka colloquially mp4)
[23:30:18 CET] <JEEB> for mov it's likely similar, but one would have to confirm @ qtff.pdf that apple gives out
[23:30:40 CET] <Jan-> Ultimately though, what this guy was saying has to be BS
[23:30:43 CET] <JEEB> (the useful part of that post is the 1/2/3)
[23:31:07 CET] <Jan-> is "box" in a MOV like "chunk" in a RIFF AVI?
[23:31:10 CET] <JEEB> so you have width/height in tkhd, pixel/sample aspect ratio in pasp, and then the video stream can have it
[23:31:29 CET] <JEEB> Jan-: I recommend building L-SMASH and testing out its boxdumper tool
[23:31:33 CET] <Jan-> these files are apple prores, so the spec for that isn't public and we don't really know if it has pixel aspect.
[23:31:35 CET] <JEEB> `boxdumper --box file.mov`
[23:31:47 CET] <JEEB> Jan-: well the mov container is what's it in
[23:31:48 CET] <furq> yeah i just checked boxdumper
[23:31:53 CET] <furq> looks like ffmpeg sets it in tkhd
[23:32:00 CET] <JEEB> yeah, tkhd is uber alles
[23:32:13 CET] <JEEB> pasp is just if you want to override what's in the video stream
[23:32:22 CET] <JEEB> and yes, tkhd is likely to be the same in mov
[23:32:34 CET] <JEEB> anyways, apple does give out mov specs and that's what really matters
[23:32:44 CET] <JEEB> that's the QTFF spec
[23:32:48 CET] <JEEB> (QT File Format)
[23:35:51 CET] <Jan-> hm these files out of this hardware recorder do have a "pasp"
[23:37:13 CET] <Jan-> and it is set to 1, 1
[23:37:34 CET] <Jan-> ehh I wonder if I can hack it in the hex editor
[23:37:47 CET] <JEEB> pasp is a funny thing because it in most cases gets overridden by tkhd's width/height
[23:37:54 CET] <JEEB> I mean, if they differ
[23:38:08 CET] <JEEB> usually tkhd's width/height are set to the presentation size
[23:38:34 CET] <Jan-> tkhd is a bigger structure
[23:38:50 CET] <JEEB> true
[23:38:56 CET] <Jan-> I'm not sure where the PAR/DAR stuff is
[23:39:12 CET] <JEEB> it's just width/height in tkhd
[23:39:23 CET] <JEEB> which is post-AR-applied
[23:39:58 CET] <JEEB> anyways, poking the pasp can lead to interesting results
[23:40:19 CET] <JEEB> although most things I know in case of tkhd being there would then just present it according to tkhd :)
[23:41:22 CET] <Jan-> you'd expect there to be a four byte long "3840" and a four byte long "2160" in a UHD file.
[23:41:24 CET] <Jan-> in tkhd
[23:41:26 CET] <Jan-> but I can't find one
[23:41:54 CET] <furq> there should be width and height keys
[23:42:16 CET] <Jan-> actually I guess this may not be the picture track
[23:43:27 CET] <Jan-> does anyone know the layout of tkhd?
[23:43:51 CET] <JEEB> just grab a copy of either qtff or isobmff spec and ctrl+f for tkhd ;)
[23:44:13 CET] <JEEB> http://standards.iso.org/ittf/PubliclyAvailableStandards/index.html
[23:44:16 CET] <JEEB> 14496-12
[23:44:19 CET] <JEEB> for isobmff
[23:45:32 CET] <Jan-> it's the last two u32s
[23:45:54 CET] <Jan-> wouldn't it be wierd to munge the track dimensions to change DAR though?
[23:46:09 CET] <JEEB> not really
[23:46:17 CET] <JEEB> that's how it's done in ISOBMFF
[23:46:34 CET] <JEEB> track width/height are with aspect ratio applied, after all
[23:46:42 CET] <JEEB> not coded dimensions
[23:46:44 CET] <JEEB> :D
[23:47:30 CET] <Jan-> okay.
[23:47:43 CET] <Jan-> It's a bit odd, I can't find the actual dimensions in tkhd
[23:52:27 CET] <Jan-> hmm. For film and TV work it might make sense to use pasp rather than tkhd
[23:52:56 CET] <Jan-> theoretically tkhd should make quicktime scale the video track before it gets presented to the application that wants the video.
[23:52:58 CET] <Jan-> this is possibly not great.
[23:54:25 CET] <JEEB> well, what would the thing show if it had a pasp of X:Y and tkhd values according to the actual sample counts?
[23:54:29 CET] <JEEB> if that's what you mean
[23:54:43 CET] <JEEB> if I recall correctly that will just end up being shown as if it's a 1:1 SAR
[23:55:20 CET] <Jan-> I'm not sure.
[23:55:29 CET] <Jan-> I have discovered that ffmpeg sets pasp when you use -aspect.
[23:55:36 CET] <Jan-> at least, there is a pasp box and its values are certainly not 1:1
[23:55:41 CET] <JEEB> with the caveat of course that I've mostly looked and worked with ISOBMFF, not QTFF. they can have minor differences :P
[23:55:48 CET] <JEEB> Jan-: check the values in tkhd as well :P
[23:55:53 CET] <Jan-> I will.
[23:56:04 CET] <JEEB> boxdumper should tell you that stuff
[23:56:29 CET] <Jan-> Problem is that I can't find sensible values in tkhd at all.
[23:56:54 CET] <JEEB> even with boxdumper?
[23:57:00 CET] <Jan-> I'm pretty sure I'm looking at the right tkhd as it has "apch" in it and that's the ProRes fourcc
[23:57:13 CET] <Jan-> well I'm just using a hex editor
[23:57:59 CET] <JEEB> feed the output of boxdumper --box to less or so
[23:58:15 CET] <Jan-> I'm not sure I can, on windows
[23:58:33 CET] <JEEB> in that case you can dump into a file with > file.txt
[23:59:06 CET] <JEEB> ok, the proprietary prores sample I have does indeed contain 720x480 in the tkhd
[23:59:12 CET] <JEEB> and 40:33 in pasp
[23:59:45 CET] <Jan-> I guess they're "longints" in the language of this hex editor
[00:00:00 CET] --- Thu Nov 24 2016