[Ffmpeg-devel-irc] ffmpeg.log.20161110

Fri Nov 11 03:05:02 EET 2016

[00:04:02 CET] <Phi> so there's no Windows Intel QSV examples?
[02:00:28 CET] <Phi> pandas
[02:00:49 CET] <Phi> I mean, I'm getting "avcodec_open2 failed with error -40: Function not implemented." for Intel QSV
[02:16:39 CET] <Phi> sempai
[04:28:09 CET] <Orbstheorem> Hello, I'm trying to apply a bandpass filter on a streamed video. I'd like to build a format-independent command line. Is there anyway to specify the output format to be equal to the input format?
[04:28:23 CET] <Orbstheorem> my cmd is youtube-dl -o - "$(xclip -o -selection clipboard)" | ffmpeg -i - -af "highpass=f=200, lowpass=f=3000" pipe:1 | mpv
[04:53:09 CET] <c_14> No
[04:53:15 CET] <c_14> (t without scripting)
[04:53:24 CET] <c_14> Though you'll want to add -c:v copy so the video doesn't get reencoded
[04:53:40 CET] <c_14> And if you don't care (much) about format support, just use matroska as that'll support everything
[04:53:59 CET] <c_14> And if you really just want to play the video in mpv you can tell mpv to apply a bandpass without piping it through ffmpeg
[04:57:41 CET] <c_14> Ah, he's not here anymore.
[04:57:43 CET] <c_14> woops
[07:04:07 CET] <Dresk|Dev> So what kind of performance can we expect on modern Android devices running 1080p?  For example, the NVIDIA Shield Tablet runs the Tegra K1, nowhere near as the X1 (which the Shield TV runs and the Pixel C uses minus 4 cores), should ffmpeg be able to playback a 1080p H.264 video at 60FPS no issues?
[07:04:39 CET] <Dresk|Dev> I've been trying so darn hard to see if ffmpeg is feasible on Android, because our software is multiplatform and we use ffmpeg to draw video to arbitrary surfaces in OpenGL
[07:04:57 CET] <Dresk|Dev> But it just seems like without hardware decoding we can never do that on Android
[07:06:01 CET] <Dresk|Dev> Does anyone have tips for what options I should for x264?  What profile?  Any optimizations when building ffmpeg for Android?  We've got NEON enabled
[07:06:55 CET] <Dresk|Dev> We're also using the latest NDK, 13b, which now uses clang over gcc/g++
[11:10:54 CET] <Mavrik> Dresk|Dev, you should always use HW decoders on those devices
[11:11:08 CET] <Mavrik> running ffmpeg is just a terrible waste of CPU cycles and will make your software less capable
[11:11:21 CET] <Mavrik> (Shield TV can easily do 4K 60fps H.265)
[12:35:35 CET] <Pandela> Morning
[12:36:31 CET] <Pandela> Anyone know if ffplay supports raw videos with audio?
[12:36:48 CET] <Pandela> I'm using "-f rawvideo -pixel_format yuv422p -video_size 1280x720 -framerate 30 -infbuf -i" But no audio is coming through
[12:39:34 CET] <BtbN> rawvideo does not have any audio. It's just a plain file with one raw frame after the other.
[12:40:44 CET] <Pandela> Darn. Thanks for the clarification.
[12:41:29 CET] <BtbN> It's nothing thar ffplay of ffmpeg don't support. It's simply nothing there to support.
[12:44:02 CET] <Betablocker> is there a easy move fading 2 rtsp streams to one ?
[12:44:31 CET] <Betablocker> i mean live streams&. not vod
[12:45:06 CET] <Pandela> BtbN: Also good to know, thanks.
[12:45:52 CET] <Betablocker> stream 1 is running &. stream 2 should fade in and replace stream 1 after eg 5 seconds
[12:46:56 CET] <Pandela> One more for tonight. Are there any other formats that support the "-pixel_format" option in ffplay/ffmpeg besides rawvideo?
[12:56:31 CET] <kerio> i'd guess that most other formats have an indicator of what the pixel format is already
[12:56:45 CET] <kerio> what do you think raw_video is?
[13:15:49 CET] <Pandela> I'm just trying to force incorrect color formats and palettes for eye candy, was curious if its possible to do with other formats
[13:23:30 CET] <kerio> oh hm
[13:23:47 CET] <kerio> i wonder if there's a way to override the pixel format
[13:23:59 CET] <kerio> (obviously, between compatible bpp formats)
[13:31:53 CET] <Pandela> That would be grand haha
[13:33:03 CET] <Pandela> Perhaps I should ask in the dev channel too eh?
[13:55:22 CET] <utack> i am guessing this is dead for forever? https://trac.ffmpeg.org/ticket/1885
[13:55:30 CET] <utack> VC1 multithreaded decoding
[14:05:38 CET] <allx> hi there - is there a limit to the length of a filtergraph string that can be passed via -filter_complex?
[14:06:38 CET] <jkqxz> utack:  Yeah, very unlikely anyone would bother.  Pretty much all modern hardware contains fast VC-1 decode support anyway.
[14:06:43 CET] <kepstin> allx: probably limited only by the max length of a command line on your system
[14:07:41 CET] <kepstin> allx: if you need a really long one, well, that's what -filter_complex_script is for :)
[14:07:54 CET] <furq> i've never seen a system with ARG_MAX less than 32k
[14:09:04 CET] <allx> kepstin, furq: great, thank you both
[14:09:13 CET] <furq> it's between 256k and 2MB on all the systems i have running atm
[14:09:17 CET] <furq> so you'll probably be fine
[14:40:30 CET] <utack> jkqxz my phone unfortunately not so much
[14:40:48 CET] <ciawal> ok I think I must have hit a bug in ffmpeg
[14:40:48 CET] <ciawal> if I switch back to 3.0.2 it sees all 69 frames of my video file, but on the latest (3.2) it sees only 68
[14:45:33 CET] <DHE> don't suppose you could bisect that...
[14:46:48 CET] <ciawal> I can try
[15:07:43 CET] <ciawal> DHE: here is the debug output with both versions https://gist.github.com/ciaran/1d6a043d13c205f377debeb550eb90b7
[15:07:46 CET] <ciawal> maybe this gives some insight
[15:08:09 CET] <DHE> do you know what bisecting is? I suspect not...
[15:08:50 CET] <ciawal> I do
[15:08:53 CET] <ciawal> it's compiling
[15:09:08 CET] <DHE> ah...
[15:09:18 CET] <ciawal> it will take some time to do
[15:43:52 CET] <zamba> i want to concatenate a four MP4 files (gopro video) and at the same time flip these
[15:43:56 CET] <zamba> how do i do that?
[15:46:58 CET] <kerio> flip them how
[15:48:49 CET] <DHE> recorded upsidedown on the gopro? sounds like you want the concat input type and then use a simple video filter to flip them back
[15:49:08 CET] <furq> if you're using the transpose filter then you might as well use the concat filter
[15:49:41 CET] <furq> or use the concat demuxer and the rotate metadata to do it losslessly
[15:50:11 CET] <kerio> furq: a better question is
[15:50:20 CET] <kerio> is it possible to losslessly rotate some but not others?
[15:50:26 CET] <furq> some what
[15:50:32 CET] <kerio> some of the input files
[15:50:46 CET] <kerio> or is the rotate metadata per-stream
[15:51:10 CET] <furq> it's set in the container
[15:51:34 CET] <BtbN> remux to ts, concat using concat demuxer and -c copy
[15:51:35 CET] <furq> if half your files are actually upside down (e.g. not rotated by metadata) then you're screwed
[15:51:42 CET] <BtbN> and then figure out how to flit it without re-encoding
[15:51:51 CET] <BtbN> *p
[15:51:56 CET] <furq> unless the inputs are jpg or something which you can actually rotate losslessly
[15:51:59 CET] <furq> although idk if ffmpeg can do that
[15:55:49 CET] <kerio> is h264 not losslessly rotable?
[15:56:27 CET] <BtbN> I suppose in theory it might be, if someone came up with logic that rotates all the motion vectors and everything else that's involved.
[15:56:35 CET] <BtbN> But such a thing for sure doesn't exist.
[15:57:41 CET] <furq> mjpeg is the only codec i know of where it's possible in practice
[16:12:05 CET] <kerio> BtbN: can't we just put a - in front of everything :D
[16:20:20 CET] <zamba> so, what's the command line option to do this?
[16:21:59 CET] <furq> depends what you mean by flip and why they're currently flipped
[16:23:49 CET] <furq> pastebin the output of `ffprobe -show_entries stream_tags=rotate input.mp4`
[16:41:53 CET] <Pandela> Anyone know if the pixel_format option in the "image2" demuxer allows you to force any colorspace?
[16:46:51 CET] <dragmore> hi, im trying to to a TS to TS remultiplex, and remove 3 out of 4 DVB-Sub pids as well as set the new language tag on the pid.. Anyone care to share how the syntax would be?
[16:47:04 CET] <dragmore> ffmpeg -i org.ts -c:v copy -c:a copy -c:s copy -map 0:v -map 0:a -map 0:s -y test2.ts
[16:47:21 CET] <dragmore> but id like to use the stream_id command and use stream 3
[16:48:31 CET] <dragmore> ffmpeg version 3.0.4
[16:48:59 CET] <furq> dragmore: https://trac.ffmpeg.org/wiki/Map#Example8
[16:50:48 CET] <zamba> furq: unrecognized option show_entries
[16:51:04 CET] <zamba> oh.. nevermind.. ffprobe, not ffmpeg
[16:51:25 CET] <zamba> http://pastebin.com/jM1hgXcg
[16:53:27 CET] <dragmore> furq: yep, but it doesn say how to re-stamp the language string
[16:53:43 CET] <dragmore> furq: today its unknown, i need to set ENG
[16:53:52 CET] <c_14> -metadata:i:id language=eng
[16:57:27 CET] <furq> zamba: so what do you mean by flip
[16:58:15 CET] <dragmore> c_14: So.. if i understand this correctly : -metadata:0x1 language=nor  ?
[16:58:39 CET] <furq> zamba: https://ffmpeg.org/ffmpeg-filters.html#concat and https://ffmpeg.org/ffmpeg-filters.html#transpose
[16:59:11 CET] <furq> you can either reencode the video the right way up, or use the concat demuxer and set the rotation metadata, which will play it back rotated in (some) players
[16:59:22 CET] <furq> the second way is much faster but won't work in all players
[17:00:22 CET] <c_14> dragmore: -metadata:i:0x1 language=nor
[17:01:32 CET] <dragmore> c_14_ ok let me try
[17:02:47 CET] <dragmore> c_14 : got a : Invalid metadata type 0.
[17:02:54 CET] <dragmore> ffmpeg -i org.ts -c:v copy -c:a copy -c:s copy -metadata:0:3 language=eng -map 0:v -map 0:a -map 0:s -y test3.ts
[17:03:30 CET] <dragmore> havnt figured out how to only copy 1 of 4 subs..
[17:04:16 CET] <furq> -map 0:s:0
[17:04:24 CET] <furq> replace the second 0 as appropriate
[17:05:23 CET] <ninja6o4> hey guys. i'm having trouble converting a really old .mov and i'm not sure if it's copy protected. is there any way to tell?
[17:05:45 CET] <dragmore> furq: still getting invalid metadata type 0
[17:05:51 CET] <dragmore> ffmpeg -i org.ts -c:v copy -c:a copy -c:s copy -metadata:0:3 language=eng -map 0:v -map 0:a -map 0:s:3 -y test3.ts
[17:10:34 CET] <ninja6o4> ffmpeg detects it as svq3 encoded
[17:11:21 CET] <ninja6o4> if anyone has a suggestion, please, i'm all ears. here's the output: http://pastebin.com/dA9x9prZ
[18:03:54 CET] <dragmore88> anyone know if ffmpeg can fix malformed PES packets for subtitles ?  ([mpegts @ 0x20a8d00] PES packet size mismatchme=00:18:51.80 bitrat&)
[18:31:35 CET] <ciawal> DHE: ca6cae73db207f17a0d5507609de12842d8f0ca3
[18:31:38 CET] <ciawal> finally
[18:50:29 CET] <Croepha> so, im using the ffmpeg api to download from an HLS source, and I only want to play one variant, (ie one video stream and its related audio stream) but the hls demuxer seems to be downloading all the streams, (all quality levels) is there a way to tell it to only download the streams I am decoding?
[19:08:36 CET] <kerio> well, the variant is just another hls stream
[19:13:37 CET] <Croepha> kerio: so you mean, just pass one of the variant playlists to avformat_open_input, instead of the main one?
[19:16:51 CET] <DHE> ciawal: interesting, and unfortunately a rather large commit...
[19:17:36 CET] <DHE> well, not so bad if you ignore the fate tests
[19:54:26 CET] <Phi> alright, an easier one for you guys
[19:54:38 CET] <Phi> "--enable-debug" seems to have options, but I can't find a list of them
[19:55:46 CET] <furq> Phi: https://gcc.gnu.org/onlinedocs/gcc/Debugging-Options.html
[19:56:03 CET] <Phi> ah, I'm building with MSVC
[19:56:24 CET] <furq> https://github.com/FFmpeg/FFmpeg/blob/master/configure#L6042
[19:56:52 CET] <furq> $debuglevel is the args to --enable-debug
[19:57:33 CET] <Phi> hmm, so it's geared for GCC?
[19:59:15 CET] <Phi> so I just have to add_cflags and add_asflags whatever MSVC needs
[19:59:59 CET] <Phi> there's --extra-cflags, --extra-cxxflags, but no asflags?
[20:01:50 CET] <furq> you can presumably use --yasmexe="yasm -args"
[20:02:21 CET] <furq> or --as rather
[20:04:48 CET] <Phi> --extra-cflags="/DEBUG" --yasmexe="yasm --DEBUG" ?
[21:03:33 CET] <Phi> --extra-cflags="--DEBUG" --yasmexe="yasm --DEBUG" seems to work
[21:03:35 CET] <Phi> ima make
[22:55:30 CET] <infinito> im trying to rely tcp to udp multicast process in ffmpeg and i get a lot of artifacts
[22:56:05 CET] <infinito> im doing that: ffmpeg  -i tcp://10.1.0.1:5000?listen -c:v libx264  -c:a copy -f mpegts   udp://239.255.255.1:6000
[22:56:52 CET] <infinito> if i check the tcp input, its comes good and its fine. the same if only i send a mp4 file to multicast
[22:57:09 CET] <infinito> but that command isnt working
[22:59:27 CET] <DHE> infinito: you didn't specify a bitrate for the video so the default is 1 megabit.
[22:59:55 CET] <DHE> for mpegts to UDP I'd recommend using full VBV mode and the output bitrate setting (assuming you're up to date with that feature)
[22:59:56 CET] <infinito> sorry, i pasted a test, instead of libx264 must be copy too
[23:00:17 CET] <DHE> ah, yes, you probably have burst bandwidth problems
[23:00:58 CET] <infinito> if i just write the tcp incoming stream to disk its works fine
[23:01:17 CET] <DHE> bandwidth burst
[23:01:19 CET] <infinito> and the multicast is in the loopback, just for make variants
[23:02:34 CET] <infinito> im now reading about vbv, its related to bandwidth burst ?
[23:02:50 CET] <DHE> well if you're using -c:v copy then it doesn't matter
[23:03:33 CET] <infinito> i am creating the stream in other place, using ffmpeg too
[23:03:38 CET] <DHE> UDP is an unreliable protocol, doubly so when going through small-buffer switches and routers. there's an option for the udp output to space the packets out by specifying the bitrate of the video.
[23:04:00 CET] <DHE> specifying slightly high a bitrate is also fine.
[23:05:24 CET] <infinito> you say i have to re encode the video to a high birtrate to make udp work fine?
[23:06:49 CET] <infinito> im  encoding the video from the tcp source  whit -b:v 2M
[23:07:31 CET] <DHE> no. I'm saying you have to tell ffmpeg what the video bitrate is so that packets are evenly spaced and match real-time flow rather than coming out at maximum speed
[23:08:48 CET] <infinito> and who can tell ffmpeg that
[23:09:07 CET] <DHE> well, you can if you set the bitrate.
[23:09:18 CET] <DHE> you might also be interested in the -muxrate parameter for the mpegts muxer
[23:10:04 CET] <furq> infinito: https://trac.ffmpeg.org/wiki/Creating%20multiple%20outputs#Teepseudo-muxer
[23:11:11 CET] <DHE> oh that's cool...
[23:12:05 CET] <infinito> furq i checked tee but i really want to use multicast to aviod rewrite code.
[23:13:31 CET] <furq> why are you using a separate ffmpeg for multicasting
[23:13:35 CET] <infinito> DHE i go to check if -f mpegts -muxrate 2M works
[23:14:05 CET] <DHE> Infiltrator: that's not good enough. the video bitrate is 2M, but the audio and muxing overhead are also present. you will get problems.
[23:14:33 CET] <infinito> furq , first i get from dvb-t in a place a lot of channels, then i send to azure trought openvpn using tcp and then, in a lot of different nodes i make variants from channels
[23:14:52 CET] <DHE> I might do something like: ffmpeg -i input.mp4 -c:b libx264 -b:v 2M -maxrate:v 2M -bufsize:v 1M -c:a copy -muxrate 2.5M -f mpegts udp://239.255.255.1:6000?bitrate=2500000
[23:15:04 CET] <DHE> oops, some typos but I think you can see what I meant
[23:15:29 CET] <infinito> yes, i check it now
[23:23:01 CET] <infinito> DHE is working!
[23:24:35 CET] <infinito> really thanks!
[23:33:12 CET] <bens218> would this be good for backing up dvd's?
[23:33:20 CET] <bens218> http://pastebin.com/jCLTERwG
[23:33:39 CET] <bens218> or is there something i should add/remove/change?
[23:34:16 CET] <llogan> 0:0 is not guiaranteed to always be a video stream. use 0:v (or 0:v:0) instead
[23:34:34 CET] <bens218> got it, thanks llogan
[23:35:16 CET] <llogan> you probably don't need -profile and -level unless you need to support some old device
[23:35:52 CET] <bens218> no, playback devices will either be kodi boxes and/or roku
[23:36:05 CET] <bens218> no old devices
[23:36:20 CET] <llogan> then you can let x264 autochoose those
[23:36:39 CET] <bens218> got it
[23:36:40 CET] <furq> one vob doesn't necessarily correspond to the thing you want
[23:36:58 CET] <furq> you probably want to use something like tccat to extract a title
[23:37:01 CET] <bens218> well, that's just an example. i know 1 single vob won't be the whole movie
[23:37:12 CET] <bens218> i've already used tccat to get my single vob
[23:37:19 CET] <furq> fair enough
[23:37:24 CET] <llogan> there is also tools/dvd2concat
[23:37:29 CET] <furq> fwiw you can pipe tccat straight into ffmpeg
[23:37:38 CET] <furq> as long as you don't care about subtitles or chapters
[23:37:48 CET] <furq> you won't get either with that command anyway
[23:37:52 CET] <bens218> i'm trying to retain the chapters if possible
[23:38:02 CET] <bens218> i can do without subs
[23:38:14 CET] <furq> i've not yet found a non-awful way to retain chapters without using mkvmerge
[23:38:31 CET] <furq> ffmpeg only supports adding chapters through some wonky ffmetadata file which nothing else supports writing
[23:38:51 CET] <bens218> oh man
[23:40:13 CET] <furq> you can still pipe tccat straight into ffmpeg anyway
[23:40:21 CET] <furq> you'll just then need to extract chapters separately and remux the output mkv
[23:40:39 CET] <furq> or write a script that converts some other chapter file format to ffmetadata
[23:41:54 CET] <bens218> would you happen to have some examples furq?
[23:45:05 CET] <furq> http://vpaste.net/aioZz
[23:45:07 CET] <furq> something like that
[23:45:37 CET] <bens218> argh, stupid work web filters. won't let me access vpaste
[23:45:53 CET] <furq> http://sprunge.us/YZbA
[23:46:26 CET] <bens218> got it, thanks
[23:47:07 CET] <furq> subtitles may or may not work with that, but last time i tried it i had to tweak analyzeduration/probesize and then the palette was messed up
[23:47:20 CET] <furq> but since you need a temp file anyway you can extract them separately if you need them
[23:48:20 CET] <ros_rox> hello - i am compiling ffmpeg on windows 8.1 using msvc++ 2008 win32, avcodec57.dll throwing entrypoint not found for calls of the kind: CoTaskMemAlloc(sizeof(VIDEOINFOHEADER)); (from pjsip library) can anyone help ?
[23:48:48 CET] <furq> also yadif=mode=1 looks much nicer than mode 0 to me
[23:49:13 CET] <furq> i don't normally use yadif for deinterlacing but frame-per-field is generally better
[00:00:00 CET] --- Fri Nov 11 2016