[Ffmpeg-devel-irc] ffmpeg.log.20180911

[burek]

[00:01:29 CEST] <hans_> [NULL @ 0000000002d1cf00] Unable to find a suitable output format for 'pipe:'
[00:01:31 CEST] <hans_> apparently not
[00:02:03 CEST] <hans_> ok, my question still stands, how do i tell it to output a jpg without trying to guess the output format from the output save name?
[00:03:05 CEST] <c_14> -f image2pipe -c:v jpeg
[00:03:06 CEST] <c_14> probably
[00:03:33 CEST] <hans_> thanks
[00:04:35 CEST] <hans_> .. why is -f image2pipe  needed?
[00:04:41 CEST] <hans_> (it is, but not sure why)
[00:05:08 CEST] <hans_> if i want to stream a video to pipe, do i need to write -f video2pipe  ?
[00:05:23 CEST] <c_14> -c:v sets the codec, but that doesn't tell ffmpeg what format to use for it
[00:05:25 CEST] <c_14> nah
[00:05:34 CEST] <c_14> you need to pick some format that's streamable
[00:05:36 CEST] <c_14> like matroska
[00:05:42 CEST] <c_14> or nut
[00:06:02 CEST] <hans_> oh lel
[00:06:29 CEST] <hans_> normally when input is a video, it stops at the first frame, but when using -f image2pipe -  , it doesn't, apparently
[00:09:50 CEST] <GuiToris> nicolas17, furq, are timestamps different from -ss/to 00:00:00?
[00:10:38 CEST] <GuiToris> I specified inpoint 3 outpoint 7 in the textfile, but the result is different from what I've expected
[00:19:13 CEST] <hans_> adding -frames 1 -r 1   made it stop after the first frame when writing to stdout too ^^
[00:39:01 CEST] <hans_> even https://www.ffmpeg.org/ffmpeg-all.html  doesn't list image2pipe
[00:39:17 CEST] <hans_> where did you learn of image2pipe ?
[00:39:54 CEST] <c_14> it might just be an alias for the image2 muxer
[00:49:07 CEST] <nurupo> is there a changelog entry for the prefixless api removal in favor of AV_ prefix in ffmpeg 4?
[00:50:05 CEST] <nurupo> i can't find it in http://git.videolan.org/?p=ffmpeg.git;a=blob;f=Changelog;h=a117ad4d2e695ce215d0092a937b885cedf3a374;hb=HEAD#l31
[01:06:18 CEST] <nicolas17> I have a folder with several videos, what's the easiest way I can get the total duration?
[01:06:42 CEST] <nicolas17> can I get ffprobe to give me the length in seconds of a video, so I can easily script the sum?
[01:09:58 CEST] <furq> nicolas17: ffprobe -v error -show_entries format=duration -of default=nk=1:nw=1 foo.mkv
[01:10:47 CEST] <nicolas17> neat
[08:46:13 CEST] <jsollie> hello everyone, does anybody know a workaround to recode the video stream of an mplayer .dump file via vaapi, copy audio streams and subtitles into mkv?
[08:46:34 CEST] <jsollie> it claims the packet headers are invalid
[08:46:49 CEST] <jsollie> is there a two-step process I could use?
[08:47:17 CEST] <jsollie> I was thinking about muxing it into m2ts, and then into mkv
[08:47:52 CEST] <jsollie> but the subtitles do not survive this, they are recognised as raw data instead
[08:49:28 CEST] <jsollie> any suggestions? I am using ffmpeg 4.0.2
[08:49:29 CEST] <Ke> you can always use mkvmerge to extract the original subtitles from original file
[08:49:46 CEST] <jsollie> yes, but there's an offset on the streams
[08:49:54 CEST] <jsollie> how do I fix that?
[08:50:08 CEST] <Ke> -y -option
[08:50:19 CEST] <Ke> for mkvmerge
[08:51:32 CEST] <jsollie> and which muxer format would you recommend as an intermediate container for the vaapi output?
[08:52:04 CEST] <Ke> no idea
[08:53:02 CEST] <jsollie> because, I just discovered, h.264 and m2ts don't seem to work well together (got this info with loglevel verbose)
[08:54:20 CEST] <jsollie> but the way you recommend it, mp4 will do fine to just recode the video stream
[08:54:28 CEST] <jsollie> which is what I want
[08:54:35 CEST] <jsollie> so let's try
[09:38:12 CEST] <jsollie> Ke: I managed to recode the video stream, but both ffprobe and mkvmerge say the video stream is invalid.  Tried this with a tonga and Fiji GPU, but the output of ffprobe is the same:
[09:38:41 CEST] <jsollie> [h264 @ 0x55bba7a1d160] separate color planes are not supported Could not open codec for input stream 0
[09:39:20 CEST] <jsollie> but aside from that, it merges the audio and subtitles with MKVmerge
[09:39:24 CEST] <jsollie> thank you!
[09:47:01 CEST] <LigH> Hi. Trying to build FFmpeg with libaom (both trunk); is there already a known issue with AOM_ERROR_RESILIENT_PARTITIONS ? I did not yet find a trac issue reported.
[18:12:41 CEST] <ahoo> your stance on shana?
[18:19:23 CEST] <satoshi2> fap
[18:30:37 CEST] <ahoo> is x265 the most sophisticated codec?
[18:33:11 CEST] <nicolas17> ahoo: probably AV1 is more sophisticated, but encoding will take several minutes per frame
[18:33:39 CEST] <DHE> I'm not sure AV1 is properly stablized yet. but h265 is
[18:34:00 CEST] <nicolas17> 'sophisticated' is pretty vague criteria to beginw ith
[18:34:06 CEST] <furq_> did the bitstream freeze actually happen
[18:34:26 CEST] <furq> i know a bitstream freeze happened but i seem to remember people in here saying they hadn't actually frozen it
[18:34:54 CEST] <DHE> last I heard it wasn't really a bitstream freeze, more of a public alpha release.
[18:35:10 CEST] <DHE> I played with it and I liked it, 25 minutes per frame at 1080p notwithstanding
[18:35:42 CEST] <nicolas17> jfc
[18:35:52 CEST] <nicolas17> does it have speed presets like x264?
[18:36:10 CEST] <furq> it's a reference codec, you're not supposed to use it
[18:36:31 CEST] <furq> i know they're planning on making that the actual recommended encoder but it isn't that now
[18:37:01 CEST] <furq> you can use rav1e if you want something usable but it's not going to be as efficient
[18:37:26 CEST] <nicolas17> video encoding is fascinating
[18:37:29 CEST] Action: nicolas17 wants to write an mpeg1 decoder some day
[18:39:41 CEST] <BtbN> It was frozen like two days ago I think
[18:58:33 CEST] <JEEB> DHE: av1 bit stream froze a few months ago I think
[18:58:36 CEST] <JEEB> june/july or so
[18:58:48 CEST] <JEEB> and the ISOBMFF (mp4) mapping got a 1.0.0 ~4 days ago or so
[19:58:47 CEST] <frendShot> If I have a bunch of files with both video and audio streams, how do I strip everything from them except the one video stream, and the English audio stream?
[20:09:24 CEST] <Hello71> man ffmpeg
[20:09:27 CEST] <Hello71> /-map
[20:13:01 CEST] <frendShot> yes, but what -map option do I use to specify a specific *language*? because not all the videos have the desired language at the same stream number
[20:14:21 CEST] <GuiToris> hey, how should I get the right 'in/outpoint' value? this concate demux behaves strangely. First it complains about the filename: "Unsafe file name './test.MTS'" (<this should be safe since it's relative) and inpoint 9 outpoint 13 is not perfectly -ss 9 -to 13
[20:15:36 CEST] <GuiToris> I've read something about intra frame codecs but I don't know how I should solve this problem
[20:16:39 CEST] <GuiToris> am I better off using -ss -to -t?
[20:27:36 CEST] <CrystalMath> hi all, i'm having a problem with -vf reverse
[20:28:01 CEST] <CrystalMath> it just uses up all my RAM and then dies because it runs out
[20:28:56 CEST] <JEEB> yea, the documentation says it does buffer the whole thing and then returns it in reverse
[20:29:09 CEST] <JEEB> because you generally cannot easily play a thing in reverse
[20:29:29 CEST] <JEEB> since the video formats in general are optimized for A->B
[20:29:44 CEST] <CrystalMath> so is there any way to make it do it slower but without dying?
[20:29:56 CEST] <JEEB> depending on your speed requirements you might want to look into vapoursynth+ffms2
[20:30:25 CEST] <JEEB> that will just keep on seeking and decoding A->B from the previous random access point, but I guess it shouldn't run out of RAM :P
[20:32:28 CEST] <CrystalMath> JEEB: does that work for video?
[20:32:46 CEST] <JEEB> yes?
[20:33:09 CEST] <GuiToris> JEEB, hi, do you happen to know what I do wrong with the concat demuxer? these inpoint outpoint directives don't work properly
[20:33:15 CEST] <JEEB> but it only works for files of course, not for streams. but trying to go B->A on a live stream or so is idiotic anyways :P
[20:33:37 CEST] <JEEB> GuiToris: it has probably o9k limitations so rather utilize the concat filter or so :P
[20:33:46 CEST] <JEEB> no idea which limitations you're hitting
[20:34:13 CEST] <JEEB> if you don't need re-encoding then just make sure you have the same PIDs for all streams and just `cat X.ts Y.ts Z.ts > concat.ts`
[20:34:37 CEST] <ChocolateArmpits> CrystalMath, you could output the video into an image sequence, rename the files in reverse order and then read and convert them to a video
[20:34:59 CEST] <GuiToris> JEEB, I need to reencode the videos, they are MTS originally
[20:35:02 CEST] <ChocolateArmpits> if you have no time to investigate what jeeb suggested
[20:35:27 CEST] <CrystalMath> ChocolateArmpits: that will only create a few million images :P
[20:35:42 CEST] <kepstin> that's basically bufferring the decoded video to disk instead of ram, since disk is usually bigger/cheaper
[20:36:01 CEST] <GuiToris> JEEB, should I deal with these? "Non-monotonous DTS in output stream 0:1; previous: 364141, current: 352258; changing to 364142. This may result in incorrect timestamps in the output file"
[20:36:09 CEST] <JEEB> yea, better to go with something that indexes the input file(s) and then going B->A . still slow but (´4@)
[20:36:32 CEST] <ChocolateArmpits> CrystalMath, is your video reall over 10 hours long ?
[20:36:35 CEST] <ChocolateArmpits> really*
[20:36:37 CEST] <kepstin> (an ideal solution is an application that decodes a gop at a time starting at the end of the file, reverses and outputs the gop, then seeks back a gop and repeats)
[20:36:44 CEST] <JEEB> GuiToris: your DTS is f'd, have fun
[20:36:52 CEST] <CrystalMath> ChocolateArmpits: no, only 1h, so ~100,000 images
[20:37:06 CEST] <GuiToris> JEEB, great, my camera does that
[20:37:15 CEST] <ChocolateArmpits> that's pretty far from the "several trillion images" figure you gave before
[20:37:17 CEST] <GuiToris> anyway, thank you for your help  :)
[20:39:09 CEST] <CrystalMath> ChocolateArmpits: i said million
[20:39:22 CEST] <CrystalMath> ChocolateArmpits: i'm sorry
[20:39:44 CEST] <CrystalMath> maybe it's doable
[20:40:18 CEST] <ChocolateArmpits> well both several million and several trillion are unfeasible to store as images, but the 100 thousand figure is surely there among possible options
[20:40:52 CEST] <ChocolateArmpits> if you're going to repeat this later on, then maybe go for vapoursynth
[20:41:21 CEST] <ChocolateArmpits> or avisynth http://avisynth.nl/index.php/Reverse
[20:41:46 CEST] <ChocolateArmpits> same function in vapoursynth http://www.vapoursynth.com/doc/functions/reverse.html
[21:00:21 CEST] <CrystalMath> ChocolateArmpits: it's working :) thank you :)
[21:00:39 CEST] <ChocolateArmpits> the image sequence suggestion?
[21:00:42 CEST] <CrystalMath> the slowest part is reassembling the images
[21:00:44 CEST] <CrystalMath> yes
[21:00:50 CEST] <ChocolateArmpits> cool
[21:01:02 CEST] <CrystalMath> i sent cat `ls * | tac` to ffmpeg
[21:01:17 CEST] <CrystalMath> the images are just 2.5 GB
[21:01:29 CEST] <ChocolateArmpits> png or jpg/
[21:01:29 CEST] <CrystalMath> because they're JPEG (i don't mind the quality loss)
[21:01:30 CEST] <ChocolateArmpits> ?
[21:01:33 CEST] <ChocolateArmpits> oh ok
[21:01:43 CEST] <CrystalMath> the video is 480p
[21:02:21 CEST] <frendShot> If I have a bunch of files with both video and audio streams, how do I strip everything from them except the one video stream, and the English audio stream?
[21:51:56 CEST] <nicolas17> CrystalMath, JEEB: in theory the most efficient way to reverse a video is one GOP at a time, rather than buffering *all of it* in RAM, but ffmpeg doesn't support that
[21:53:08 CEST] <JEEB> yes
[21:53:13 CEST] <JEEB> because that's a very specific feature
[21:53:30 CEST] <JEEB> also it requires you to be able to quicly move to the previous GOP's RAP
[21:53:37 CEST] <JEEB> so something like ffms2's index would make it possible
[22:26:59 CEST] <kepstin> for a file with a seek index that ffmpeg knows how to use, building this using ffmpeg's seek api doesn't seem unreasonable imo.
[22:27:13 CEST] <kepstin> iirc ffms2's index is mostly to map frame number to pts?
[22:28:23 CEST] <marcurling> Hello, using ffmpeg-10bit, what is the correct (ouput) option to convert a 10bit file to 8bit, please?
[22:29:26 CEST] <JEEB> there is no such thing as ffmpeg-10bit
[22:29:31 CEST] <JEEB> that clearly means something very specific
[22:31:54 CEST] <kepstin> if you're using an old build of ffmpeg statically linked to a 10-bit-only build of x264, the correct option is to throw the binary in the trash and get a new one
[22:32:30 CEST] <kepstin> (that binary would not support outputting 8-bit h264 no matter what options you use)
[22:32:39 CEST] <marcurling> JEEB: when I say ffmpeg-10bit, I mean a compiled version of ffmpeg made with -depth 10 (or something like that) ; well, able to encode in 10 bit
[22:33:21 CEST] <durandal_1707> marcurling: you use some very special version? is that for x264 encoder?
[22:33:25 CEST] <JEEB> marcurling: FFmpeg itself isn't limited, most likely it is what kepstin says and you have a 10bit-only build of libx264 linked against it
[22:33:37 CEST] <JEEB> in which case you do not get 8bit with taht build
[22:33:56 CEST] <JEEB> libx264 can now output both 10bit and 8bit with a single library, so a newer FFmpeg with a newer libx264 should do it
[22:33:59 CEST] <JEEB> :P
[22:37:22 CEST] <marcurling> Got it. Thanks
[23:21:01 CEST] <rjeli> how does 10 bit work?
[23:21:08 CEST] <rjeli> do you put each pixel in a 32 bit word?
[23:21:47 CEST] <rjeli> oh its part of the encoding spec
[23:31:46 CEST] <kepstin> rjeli: if you're curious about how decoded video is represented in memory - one common way in unpacked planar video is to use 16bit ints per component, of which only 10 bits are used
[23:32:04 CEST] <kepstin> there's more memory-efficient packed formats as well, but they're harder to work with, so... all trade-offs.
[23:34:01 CEST] <rjeli> gotcha thx
[00:00:00 CEST] --- Wed Sep 12 2018