[Ffmpeg-devel-irc] ffmpeg.log.20150211

burek burek021 at gmail.com
Thu Feb 12 02:05:01 CET 2015

[00:54] <voip_> guys,  how to send HLS live stream to "localhost"? Than i need to pull it from other server
[00:54] <voip_> ?
[05:46] <YaMoonSun> How do I copy a specific audio stream? I tried -acodec: copy but it always grabs thre first audio stream, and I need japanese.
[05:48] <relaxed> YaMoonSun: read about -map
[05:49] <relaxed> https://trac.ffmpeg.org/wiki/How%20to%20use%20-map%20option
[05:49] <YaMoonSun> I also included -map 2 and tried even -map:2 and they didn't work for me
[05:49] <YaMoonSun> Will do though, thanks
[05:58] <YaMoonSun> I've half a mind to cheat and use handbrake
[06:03] <relaxed> YaMoonSun: pastebin the output of ffmpeg -i yourinput
[06:05] <YaMoonSun> http://pastebin.com/NuwqtWXV
[06:06] <relaxed> you want video, jpn audio, and subs in the output?
[06:07] <relaxed> ffmpeg -i input -map 0:v -map 0:a:1 -map 0:s -c:v -c:a copy -c:s copy output
[06:09] <relaxed> -map 0:a == all audio from the first input; -map 0:a:0 == first audio stream; -map 0:a:1 == second (jpn) audio stream
[06:11] <relaxed> (that should have been "-c:v copy")
[06:15] <YaMoonSun> Pfft, sounds difficult, will give it a shot though, thanks.
[06:26] <YaMoonSun> "At least one output file must be specified"
[06:27] <relaxed> output was a generic name, use output.mkv
[06:28] <YaMoonSun> Using "InuYasha S01E01.mkv" as output
[06:28] <YaMoonSun> Still not working
[06:28] <relaxed> show me the command
[06:31] <YaMoonSun> ffmpeg -i "input.mkv" -map 0:v -map 2:a -vf subtitles=input.mkv -vf scale=640:480 -c:v libx264 -b:v 1M -c:a copy -threads:4 "Output.mkv"
[06:33] <relaxed> ffmpeg -i "input.mkv" -map 0:v -map 0:a:1 -map 0:s -vf scale=640:480,subtitles=input.mkv -c:v libx264 -b:v 1M -c:a copy -threads 4 "Output.mkv"
[06:34] <relaxed> oh wait, omit -map 0:s if you're hardsubbing
[06:37] <YaMoonSun> I think it just hates me honestly.
[06:38] <YaMoonSun> Even if I manage, what are the chances that the subtitles will be scaled with the rest of the video?
[06:43] <relaxed> they should look fine. If you're still getting an error then pastebin your command and output.
[06:55] <YaMoonSun> http://pastebin.com/z1sQHpwG
[07:04] <relaxed> YaMoonSun: it says only text based subtitles are supported.
[07:05] <YaMoonSun> Managed to get it to work with handbrake, and handbrake using ffmpeg in the background.
[07:05] <relaxed> ok
[07:05] <relaxed> https://trac.ffmpeg.org/wiki/HowToBurnSubtitlesIntoVideo
[07:12] <YaMoonSun> http://pastebin.com/ZJ4wmGph
[07:15] <relaxed> YaMoonSun: remove -map 0:s
[07:17] <relaxed> ffmpeg -i "title03.mkv" -filter_complex "[0:v][0:s]overlay[v]" -map "[v]" -map 0:a -c:v libx264 -b:v 1M -c:a copy -threads 4 "InuYasha S01E01.mkv"
[07:17] <YaMoonSun> Same error message, not connected to any destination
[07:20] <relaxed> oops, that should have been -map 0:a:1
[07:24] <YaMoonSun> This program is quite the headache at times.. Thanks for helping out though
[07:25] <YaMoonSun> Does x264 automatically use a vbr?
[07:29] <relaxed> You probably want -crf instead of -b:v, see https://trac.ffmpeg.org/wiki/Encode/H.264
[07:31] <YaMoonSun> Wasn't bad, but the audio is english, going to keep playing with it.
[07:31] <relaxed> ffmpeg will do exactly what you tell it to do, I guess the headache is learning what to tell it
[07:32] <relaxed> you need to change "-map 0:a" to "-map 0:a:1" in the above command
[07:33] <relaxed> you want to encode a small sample to check, add -t 60 (seconds)
[07:33] <relaxed> er, if you want to...
[07:36] <YaMoonSun> Can I not rescale to 640:480?
[07:48] <relaxed> YaMoonSun: try, ffmpeg -i "title03.mkv" -filter_complex "[0:v][0:s]overlay[tmp];[tmp]scale=w=640:h=480[v]" -map "[v]" -map 0:a:1 -c:v libx264 -crf 18 -c:a copy "InuYasha S01E01.mkv"
[07:56] <YaMoonSun> I think I'm going to have to take a break honestly, I'm getting quite frustrated.
[10:39] <k_sze> We seem to be running into a problem where seeking a .nut file that is still being written is unreliable.
[10:40] <k_sze> s/written/recorded into/
[10:45] <khali> with the latest ffmpeg I get a strange message on the 2nd pass of 2-pass encoding:
[10:45] <khali> IO error: Success
[10:45] <khali> is this a known issue?
[10:50] <relaxed> does it happen with git?
[11:07] <khali> relaxed: yes, that's what I'm using
[11:08] <khali> does not happen with 2.3.6 as packaged in my distro
[11:08] <khali> relaxed: bisection in progress
[11:14] <relaxed> khali: ok, is it happening with a specific encoder?
[11:47] <khali> relaxed: I'm encoding with libxvid + libmp3lame
[11:48] <khali> relaxed: no problem with libx264 + libfaac but this is a one-pass encoding so that might matter too
[13:00] <khali> relaxed: doh, my bisection is probably all wrong
[13:00] <khali> ffmpeg binary got renamed to avconv somewhere in the middle, but I was always calling ffmpeg
[13:01] <BtbN> uhm, are you sure you are bisecting the right project? Or did you get caught up in some merge from libav?
[13:05] <khali> BtbN: I'm using git://source.ffmpeg.org/ffmpeg
[13:07] <khali> doh, must go to school fetch my kids
[13:07] <khali> brb
[13:50] <_Vi> Why when storing RGB24 rawvideo data to mkv files, it puts 00000000 in ColourSpace element? Such files can't be decoded by even by FFmpeg itself even if I put "-pix_fmt rgb24" explicitly.
[13:51] <_Vi> I think FFmpeg should store the string value of "-pixel_format" option instead of zeroes. Or least there should be a warning when muxing.
[13:54] <PSvils> hey guys! if I miss any important details, please let me know, I'm very new with ffmpeg! I'm encoding to VP8 and need to set a quality appropriate for realtime. I do av_opt_set(codecContext->priv_data, "quality", "realtime"), works on Windows, but fails on Android with "Option not found".
[13:54] <iive> _Vi: would you fill a bugreport at https://trac.ffmpeg.org/ ?
[13:54] <PSvils> If anyone would have any ideas as to why, that would be great! :) (sorry for wall of text)
[13:55] <_Vi> iive, OK. Shall I also implement a patch?
[13:56] <iive> _Vi: if you know how, you can send it directly to ffmpeg-dev maillist.
[13:58] <khali> relaxed, BtbN: faulty commit found, it's 934f2d2f5c16df8aad9f401a9fd842b5d9a78b11
[13:59] <PSvils> in order to set the "quality" option to "realtime", do I need to enable something when compiling ffmpeg or libvpx? (compiling to libavcodec)
[13:59] <khali> the bug is pretty obvious actually, the error message is printed on the success path too
[13:59] <khali> there's a conditional missing
[14:00] <khali> I'll write a fix, should I report on trac or send it directly to ffmpeg-devel?
[14:06] <_Vi> iive, Filed: https://trac.ffmpeg.org/ticket/4305
[14:18] <khali> patch sent
[14:24] <sheldor> Hi thanks for the patch http://git.videolan.org/?p=ffmpeg.git;a=commitdiff;h=ac494e5a66507c8124e8ba399b3f0685a51d7a82#patch1
[14:24] <sheldor> but How can I patch it, I don't have libavformat/isom.c
[14:27] <c_14> Just pull/clone the git?
[16:12] <jameshowe> follow-up from yesterday: the changes that fix --enable-pic are 581c7f0e and f963f803
[16:47] <_Vi> What codec should be used for very fast shallow compression? The output is intended to be re-compressed slowly later. Scenario: screen recording with x11grab, avoiding CPU.
[16:48] <jameshowe> raw?
[16:48] <jameshowe> i.e. no codec
[16:48] <iive> huffyuv should be fastest.
[16:49] <ramiro> iive: isn't ffv1 faster?
[16:49] <iive> no, ffv1 uses cabac and that is slower.
[16:50] <iive> mjpeg with high quant is preferable for live video, screen capture ... not so good (lot's of sharp edges).
[16:50] <iive> x264 with qp=0 is lossless and employs motion compensation, if you can afford the cpu time.
[16:51] <iive> it is slower than ffv1 (it also uses cabac), but it employs motion compensation, so it provides better compression.
[16:52] <ubitux> utvideo is another candidate
[16:53] <PSvils> hey guys! another question - what are possible reasons for having a jumpy video when encoding? my pts value is definitely increasing, albeit a bit sporadically (input stream frames are inconsistent)
[16:53] <_Vi> jameshowe, I expect some compression, for example, 2-3-wise compared to just storing yuv420p.
[16:53] <PSvils> any hints can help! :)
[16:53] <jameshowe> ah, well you didn't say that ;)
[16:53] <PSvils> (the video export seems to have frames jumping between each other)
[16:53] <_Vi> iive, The idea is to economize CPU and IO bandwidth simultaneously.
[16:53] <PSvils> it's as if every now and then, it shows a frame from a split second before.
[16:54] <_Vi> iive, Recording 60 FPS screencaps with low CPU and disk IO.
[16:54] <jameshowe> PSvils sporadic pts could do it, or poor B-frame handling
[16:55] <PSvils> jameshowe: any fixes for sporadic pts? I'm writing to webm, using 1000 fps and pts being milliseconds.
[16:56] <PSvils> (VP8 codec)
[16:57] <jameshowe> 1000 fps?
[16:57] <jameshowe> as the output framerate?
[16:57] <PSvils> well, the timebase is 1 / 1000
[16:57] <iive> _Vi: well, it is always a trade-off. you trade cpu for compression/bandwidth.
[16:58] <iive> raw is fastest, if you HDD can handle the load.
[16:58] <iive> you/your
[16:58] <_Vi> iive, rawvideo is extreme end. "libx265 -preset placebo" is other extreme end.
[16:59] <_Vi> iive, I want something between "libx264 -preset ultrafast" and "rawvideo".
[17:00] <iive> how does ultrafast -qp 0 work for you?
[17:00] <iive> you might also try some of the fraps codecs... they are meant for screen capture.
[17:01] <_Vi> iive, with ultrafast -qp 0 there is big IO margin, but one CPU core is loaded (real time == user time).
[17:03] <_Vi> Looks like there is only fraps decoder in FFmpeg, not encoder.
[17:05] <iive> hum...
[17:48] <_Vi> Which FFmpeg-supported UDP-multicast-friendly container support all/most codecs (like mkv and nut)?
[17:50] <Mavrik> define "all/most"
[17:50] <Mavrik> my far most widely used it MPEG2-TS
[17:51] <_Vi> Saving to mkv rarely fails because of unsupported codec. In particular, there is limited support of rawvideo.
[17:52] <_Vi> I want likewise universal container format, but which also likes streaming like mpegts.
[17:53] <_Vi> "ffmpeg ... -f nut udp://..." works poorly. In particular, the receiving side can't "connect" to the stream if it has missed the beginning.
[18:24] <Razzorx> hey, I'm trying to cross-compile ffmpeg with libx264 on ubuntu for arm/raspi with crosstool-ng, but it stops after telling me that it can't find pkg-config
[18:27] <Razzorx> I was wondering why it fails, the log tells me that it tries to compile files which don't exist, like a windows.h
[18:50] <AV_Damien> is there any way to set an opacity to 50% of a png for a water mark ? something like this : ffmpeg -i aa.mov -i image.png -filter_complex 'overlay=10:10, opacity=0.5' bb.mp4
[19:13] <Jan-> hihi
[19:14] <Jan-> I have some files (prores quicktime, lots of channels of audio) which won't load into an NLE. But if I convert them with -c:a copy -c:v copy they work fine. Although... what's that really changed, then?
[19:14] <c_14> Depends
[19:25] <Jan-> I guess in an ideal world I could leave out two of the four audio tracks, if there's a way to do that. They don't have any content.
[19:25] <c_14> Don't map them
[19:25] <Jan-> !?
[19:26] <c_14> If I could see your current command, I could tell you what to change.
[19:26] <Jan-> It's generally something like ffmpeg -i %%INFILE%% -c:a copy -c:v copy %%OUTFILE%%
[19:27] <Jan-> doesn't get much simpler than that
[19:27] <Jan-> theoretically they do have timecode tracks but we don't need them
[19:27] <c_14> That'll automatically throw away all but the "best" video and audio streams.
[19:28] <Jan-> There's only one video track and one audio track (with four channels of audio in it)
[19:28] <Jan-> plus the timecode
[19:28] <Jan-> Presumably there's a way of maintaining that but who knows
[19:29] <c_14> You want to throw away audio channels, use -map_channel
[19:29] <Jan-> it doesn't really matter
[19:30] <Jan-> it's just annoying when the editors drop them on a timeline and get four mono tracks
[19:30] <llogan> What NLE has trouble with your ProRes input files?
[19:30] <Jan-> Premiere
[19:30] <llogan> which version?
[19:30] <Jan-> some ancient thing
[19:30] <Jan-> it's a known issue, it doesn't like multichannel audio. except when it does.
[19:31] <llogan> sounds like a NLE
[19:31] <Jan-> I think ffmpeg is turning them from one 4 channel track into four 1 channel tracks or something
[19:31] <Jan-> but it's hard to tell
[19:31] <llogan> you need to include the output from ffmpeg for us to look at
[19:31] <Jan-> it was actually a mistake made by the camera crew, they left the recorder set to record two sdi channels and two analog channels when all they wanted was the analog.
[19:31] <Jan-> well, it's really big, I mean...
[19:32] <llogan> that's what pastebin is for
[19:32] <llogan> also tell the camera crew to stop pulling the XLR cables from the cable itself. i'm tired of resoldering them.
[19:35] <Jan-> http://pastebin.com/NY5wYLZL
[19:36] <tartinette> Hi
[19:37] <llogan> Jan-: it's outputting a single audio stream that contains 4 channels
[19:37] <llogan> do you like the Atomos?
[19:38] <llogan> Jan-: also your ffmpeg is old.
[19:40] <Jan-> llogan: It works.
[19:40] <Jan-> Don't fix it :)
[19:40] <llogan> you should always use a recent build
[19:41] <llogan> if a new build causes issues we would be really interested to know
[19:41] <Jan-> I'm afraid we're more interested in what works for us as opposed to what works for you :D
[19:42] <Jan-> we do also find that -c:v prores works whereas -c:v prores_ks causes endless grief and hassle.
[19:42] <llogan> you're missing at least 12,445 updates to FFmpeg
[19:42] <llogan> old builds are not supported here
[19:42] <Jan-> s'OK, we don't miss them.
[19:42] <Jan-> well sure, OK, but to be fair, an old build of ffmpeg is from this morning.
[19:42] <llogan> ok, then we can not provide further support
[19:44] <Jan-> well, I'm sorry, but I'm not in a position to just randomly updated and risk breaking stuff we use daily.
[19:45] <llogan> and we don't have the time or sufficient volunteers to support old stuff
[19:45] <Jan-> problem is if the new stuff breaks you're probably not really going to support that either
[19:45] <Jan-> it's OK, I don't really expect much support, it's free afterall :)
[19:45] <llogan> regressions are considered important to us
[19:47] <Jan-> I think the problems with prores encoding were why we didn't upgrade
[19:48] <Jan-> this version seemed to work best
[19:48] <Jan-> but that may be fairly unscientific
[19:48] <kepstin-laptop> well, the prores stuff in the new version won't ever get fixed if nobody tries it and tells the devs that something's wrong...
[19:49] <Jan-> it's open source, I don't think anything ever really gets fixed!
[19:49] <llogan> then don't use it and go buy something "enterprise"
[19:49] <llogan> or take a look at the bug tracker
[19:51] <Jan-> I don't really speak "bug tracker" :/
[19:52] <Jan-> actually there are now some commercial prores encoders for windows
[19:52] <llogan> my point is that bugs are fixed often
[19:52] <Jan-> but they seem to have suspiciously identical bugs to ffmpeg's :/
[19:52] <llogan> or your NLE is old and crappy no matter the input
[19:53] <__jack__> (or these soft are ffmpeg-based, which is funny)
[19:53] <flaviocpontes> I know of at least a commercial broadcast product base on FFMPEG: Newtek Tricaster.
[19:54] <Jan-> llogan: I'm sure that's true but unfortunately that's what exists.
[19:54] <Jan-> adobe are really annoying
[19:54] <flaviocpontes> I had the opportunity to operate one and it's got ffmpeg installed.
[19:54] <llogan> flaviocpontes: i'd be interested in seeing the configure line
[19:55] <Jan-> lots of things do
[19:55] <flaviocpontes> Me too.
[19:55] <Jan-> I wouldn't get too excited about people just distributing builds.
[19:55] <Jan-> Most people who actually use ffmpeg would be unable to if they could only get source
[19:56] <flaviocpontes> FFMPEG is the base for a media transcoder I'm developing for my employer
[19:56] <flaviocpontes> Even though I use them through the shell and parse the stderr output
[19:58] <Jan-> I think we've all done that
[19:58] <flaviocpontes> It's the easier way
[19:58] <Jan-> I did once try to use the avcodec/avformat APIs in C
[19:59] <Jan-> the documentation is zero
[19:59] <flaviocpontes> I couldn't find python binding for the libs
[19:59] <Jan-> I'd have liked to have p/invoked it from C#
[19:59] <flaviocpontes> My exact problem
[19:59] <Jan-> but yikes
[19:59] <Jan-> it's bad enough trying to do interop with normal windows libraries, let alone avcodec.dll (even if you can actually get it as a DLL)
[19:59] <flaviocpontes> pyMovie does the same but they don't parse the output, though
[20:00] <Jan-> I've always felt it was a pretty stupid set of rules
[20:00] <flaviocpontes> Can't say nothing about windows. All my work is in debian
[20:00] <Jan-> I mean, you can use it, but you have to use it in a really bad way.
[20:00] <Jan-> S'just stupid
[20:03] <Jan-> the normal prores encoder does seem to use heaps of bitrate.
[20:04] <Jan-> but I guess that's OK
[20:04] <stein_> hi all
[20:04] <stein_> how can i create a 25i video from a 25p source?
[20:05] <stein_> is there something like double deinterlace with full motion ?
[20:05] <Jan-> you just want to change the way it's marked in the stream, or actually make it look different?
[20:06] <stein_> make it look different .. so its playable on a tv
[20:06] <stein_> is that, what yadif=1 is for?
[20:06] <Jan-> well, 25p is playable on a TV
[20:06] <llogan> yadif is a deinterlacer, but from your question i assumed you wanted to introduce interlacing
[20:06] <kepstin-laptop> 25i isn't playable on any tv I know about; pal is 50i, ntsc is 60i(ish)
[20:16] <stein_> ah sry then its 50i .. 25p to 50i
[20:17] <Jan-> basically no.
[20:17] <Jan-> you could theoretically do it with optical flow interpolation and try to make it look like 50i
[20:17] <Jan-> but a TV will play back 25p anyway
[20:17] <Jan-> it'll just look more like a "movie" than a "tv show"
[20:19] <kepstin-laptop> well, if you actually have an old crt around, it'll scan half the rows of one frame, then the other half of the rows of the same frame, then go to the next.
[20:19] <kepstin-laptop> if you have an lcd tv, it'll probably realize that the input's really 25p and just display it that way
[20:21] <Jan-> depends how you fed the signal to it I guess
[20:21] <Jan-> does ffmpeg have any optical flow framerate interpolation stuff?
[20:21] <kepstin-laptop> not as far as I know
[20:21] <llogan> no. see slomoVideo
[20:22] <fahadash> Can I ask question related to audio editing? specifically noise-removal
[20:22] <Jan-> is there some reason for that? I'm sure there's an open source library
[20:22] Action: kepstin-laptop hates in when tvs have that, the 120hz interpolation always looks really weird :)
[20:22] <llogan> *slowmoVideo
[20:22] <stein_> ok thank you
[20:22] <stein_> guess the producing side have to change the format then
[20:22] <llogan> Jan-: nobody has submitted a patch
[20:22] <Jan-> I'm quite surprised
[20:22] <Jan-> it does everything else!
[20:23] <kepstin-laptop> well, if someone wants ffmpeg to do that, they should either write a patch or hire someone to write a patch. open-source doesn't happen by magic :)
[20:23] <stein_> isnt 25p like 50i top + 50i bottom combined?
[20:23] <stein_> or, it should be at least?
[20:23] <kepstin-laptop> stein_: in real interlaced video, the alternate rows are from different times
[20:23] <kepstin-laptop> but in progressive video, the entire frame is from the same time
[20:24] <stein_> got it
[20:25] <Jan-> kepstin-laptop: The problem is that few people have the ability to write patches, and the cost of hiring software engineers is way more than the cost of buying commercial software to do the job.
[20:26] <kepstin-laptop> Jan-: then do a crowd-funding campaign to hire a developer to add the feature :)
[20:27] <Jan-> Surely the motion compensation code is a large part of what you'd need anyway.
[20:27] <kepstin-laptop> the reason that ffmpeg can do windows screen grabbing nowadays is that I needed it for work stuff, so I dug up some old code, fixed it up, and submitted it.
[20:30] <ShadowJK> iirc, Mplayer had at one point a motion compebsation/interpolation based deinterlacer
[20:31] <Jan-> kepstin-laptop: not everyone is a software engineer.
[20:31] <kepstin-laptop> ShadowJK: yeah, 'mcdeint'; it's in ffmpeg
[20:32] <kepstin-laptop> i never really saw any improvement from it over just using yadif.
[20:33] <kepstin-laptop> (and it's hella slow, too)
[20:34] <ShadowJK> from what I remember staring at ffmpeg's motion vectors in encoded videos, their prime purpose wasnt to detect actual motion, but find most efficient way to encode the next frame
[20:45] <Jan-> maybe
[20:45] <Jan-> I'm not sure
[20:55] <anYc> Hi, I try to cut out the Ads of a TV recording in an AVI (h264,mp3) container. To cut between I frames, I recode the part to the next I frame and just copy the remaining part until the next I frame. While this works mostly, I always get small glitches between the copied and recoded part. The commands: http://pastebin.com/QMU3awYP Does anyone have a hint what I could try to debug this?
[20:55] <anYc> other sites recommend avidemux for this task but I want to get this to work with ffmpeg
[20:57] <anYc> the settings of the single files look equal according to "ffmpeg -i"
[21:00] <anYc> mediainfo shows an additional entry for the audio in the file with the copied part: "Interleave, preload duration: 24 ms". maybe this is causing the small glitches?
[21:03] <fahadash> What are noise profiles used for?
[21:03] <ganthore> fahadash: noise reduction?
[21:05] <ganthore> fahadash: oh, you mean this: https://gist.github.com/anandology/6422356
[21:06] <fahadash> Someone told me that he records sounds and each sound file comes with a noise-profile file which is another mp3 file. He said that that other file is used for noise reduction purposes
[21:17] <llogan> fahadash: you can use sox or audacity for that
[21:17] <fahadash> I already got the examples for that. I am trying to learn the basics of noise reduction using profiles. Is there any place I can read about that?
[21:18] <fahadash> I have been looking for that on Google, most of the noise reduction stuff I found had to do with videos
[21:32] <nyuszika7h> ERROR: ffprobe or avprobe not found. Please install one.
[21:32] <nyuszika7h> I'm getting this from youtube-dl
[21:32] <nyuszika7h> I actually have ffprobe
[21:33] <nyuszika7h> for some reason it's not in the PATH but ffmpeg is?
[21:33] <nyuszika7h> hmm nope, neither are, but it found ffmpeg
[21:33] <nyuszika7h> will use ff-prompt
[23:35] <pomaranc> does 'PES packet size mismatch' imply damaged input?
[00:00] --- Thu Feb 12 2015

More information about the Ffmpeg-devel-irc mailing list