[Ffmpeg-devel-irc] ffmpeg.log.20150323

burek burek021 at gmail.com
Tue Mar 24 02:05:01 CET 2015


[00:13:17 CET] <buu> Stream specifier ':1' in filtergraph description [0:1] [0:2] [1:1] [1:2] [2:0] [2:1] [3:0] [3:1] concat=n=4:v=1:a=1 [v] [a] matches no streams.
[00:13:29 CET] <buu> Which stream specifier is it talking about?
[00:22:03 CET] <buu> What could possibly be wrong with that syntax?
[00:22:28 CET] <c_14> The problem is probably that one of your inputs doesn't have a second stream
[00:23:04 CET] <buu> Every input has at least 4 streams
[00:26:10 CET] <buu> c_14: These are the files: http://paste.debian.net/162620
[00:26:16 CET] <buu> This is the command http://pastie.org/10046153
[00:27:09 CET] <c_14> What about the command output?
[00:27:21 CET] <buu> I gave you the error above
[00:27:37 CET] <buu> oh wait
[00:27:41 CET] <buu> That's not the command that errored =/
[00:27:46 CET] <buu> one sec
[00:28:11 CET] <LanDi> hwy guys
[00:28:19 CET] <LanDi> I'm trying this... ffmpeg -i video.avi -vf subtitles=subtitle.srt out.avi
[00:28:35 CET] <LanDi> but it's not working... .anyone interested in see my output ?
[00:29:01 CET] <c_14> LanDi: sure, is ffmpeg erroring out or is something not as expected?
[00:29:39 CET] <buu> c_14: I switched to the 1:v style syntax and it worked, but my original command was thus: http://paste.debian.net/162622/
[00:29:58 CET] <buu> c_14: Which produces this: http://paste.debian.net/162621
[00:30:42 CET] <LanDi> c_14: http://pastebin.geany.org/OIo1o/
[00:30:45 CET] <buu> oh
[00:31:05 CET] <buu> That needs to be [1:0] doesn't it
[00:31:37 CET] <c_14> yep
[00:31:43 CET] <c_14> also 2:0 and 3:0
[00:32:08 CET] Action: buu gives that error message a solid D-
[00:32:40 CET] <c_14> LanDi: the file why.srt doesn't exist
[00:32:45 CET] <buu> My 2: and 3: were correct, just not the 1:
[00:32:55 CET] <buu> But 1:1 does match a stream, its just not the video stream
[00:33:08 CET] <LanDi> c_14: it does... it's in the same directory
[00:33:09 CET] <c_14> oh, yeah, right
[00:33:22 CET] <c_14> LanDi: file permissions?
[00:33:37 CET] <LanDi> hmmm... let me try as root
[00:34:05 CET] <LanDi> c_14: nope, the same error
[00:38:18 CET] <c_14> What's the output of `ls -l why.mp4 why.srt' ?
[00:39:35 CET] <LanDi> c_14: http://pastebin.geany.org/pQR2Z/
[00:41:23 CET] <c_14> Why is why.mp4 empty? And why.srt looks rather small
[00:42:00 CET] <LanDi> c_14: both are small
[00:42:08 CET] <LanDi> it's just a 1:30 video
[00:46:07 CET] <LanDi> c_14: I have a new output... I have changed the srt
[00:46:33 CET] <LanDi> c_14: http://pastebin.geany.org/osVPn/
[00:47:13 CET] <pyBlob> playing a video using ffplay ... but the output range is only 16..235 and not full range, any way to switch to full range?
[00:48:32 CET] <c_14> LanDi: you can't use the same file as input and output. It breaks things
[00:48:49 CET] <LanDi> c_14: ohh. sorry
[00:50:45 CET] <LanDi> c_14: :( the same issue
[00:51:46 CET] <c_14> Yeah, because you already overwrote the input.
[00:52:13 CET] <c_14> pyBlob: try -vf scale=in_range=full
[00:53:43 CET] <pyBlob> nope, still looks the same
[01:01:54 CET] <c_14> try -color_range 2
[01:02:22 CET] <pyBlob> that makes the output even more grayier
[01:02:48 CET] <c_14> That should make it full range though
[01:06:45 CET] <pyBlob> yeah, putting -i in front of the filename and the -vf after -i WORKED =)
[01:07:09 CET] <pyBlob> ... in combination with the color_range 2
[01:13:57 CET] <pyBlob> ok ... but doesn't work with every video :(
[01:24:46 CET] <pyBlob> the problem seems to be within ffplay itself: "ffplay black.png" isn't black
[03:11:36 CET] <gdoteof_> Warning: matroska_reader: (MS compatibility mode, track 0) Matroska says video height is 1080, but the BITMAPINFOHEADER says 4294966216.
[03:12:30 CET] <gdoteof_> can i fix this?  if i play the file with ffplay -codec rawvideo -pixel_format rgb24 it is like it is off soem small number of bytes
[03:12:40 CET] <gdoteof_> tracks weird, makes weird colors
[03:52:07 CET] <Dougie187> I'm trying to create a video from a set of jpg's, and for some reason when I use the -r to set the fps to be 2 it drops a bunch of the images.
[03:52:14 CET] <Dougie187> Any idea how to prevent it from dropping the images?
[04:01:25 CET] <temhawk|afk> how do I create an animated gif made up of several animated gifs from different parts of a video arranged in a grid? like those video screenshot previews, but with animated thumbnails instead of static images
[04:11:00 CET] <Hurnhu> Hello I am making a program that uses ffmpeg. And I was wondering can Is there I can realease my program at free closed source?
[05:30:58 CET] <fffan> I have some audio frame in an buffer, but I don't know it's format, how can I guest the format and create correspondent AVCodecContext to decode.
[11:51:23 CET] <temhawk> how do I create an animated gif made up of several animated gifs from different parts of a video arranged in a grid? like those video screenshot previews, but with animated thumbnails instead of static images
[14:04:06 CET] <Infiltrator> I ran `ffmpeg -i in.mkv -q 1 out.mkv` but it's showing q=28.  Why won't it listen to me?  :(
[14:16:33 CET] <Infiltrator> Also, it segfaulted.  :(
[15:21:27 CET] <Elirips> Hello. How can I tell ffmpeg to write one frame per second of a videostream to a file, overwritting the existing file and *not* numbering files using %03d or similar?
[15:22:23 CET] <Elirips> If I try with something like ffmpeg -i <mysource> -r 1 -f image2 foo.png it complains that 'foo.png' does not containt a pattern like '%03d'
[15:26:42 CET] <temhawk> Elirips: what do you want the resulting file to look like?
[15:27:22 CET] <Elirips> temhawk: good :)
[15:27:27 CET] <Elirips> Sorry, I dont get your question?
[15:28:09 CET] <temhawk> Elirips: you want to create an image? how are you going to put multiple frames inside an image?
[15:28:22 CET] <Elirips> temhawk: the image shall only contain the last frame
[15:28:44 CET] <Elirips> but it should be updated constantely by ffmpeg
[15:28:56 CET] <temhawk> what are you trying to do?
[15:29:10 CET] <Elirips> creating thumbnails every second or so
[15:29:14 CET] <Elirips> or every minute, doesnt matter
[15:29:42 CET] <temhawk> what are you going to do with that file if ffmpeg is constantly overwriting it?
[15:30:11 CET] <Elirips> serve it using apache
[15:30:38 CET] <Elirips> without having to change the url for every image created
[15:31:00 CET] <temhawk> but the file is constantly gonna be changing
[15:31:06 CET] <Elirips> yes
[15:31:09 CET] <Elirips> thats correct
[15:31:18 CET] <Elirips> say, ffmpeg extract a frame every 10 seconds
[15:31:29 CET] <Elirips> so the users gets an updated thumbnail every 10 seconds or so
[15:31:46 CET] <temhawk> uhm, I don't think that's how that works
[15:31:51 CET] <Elirips> well, I can also define cron-jobs to run ffmpeg every 10 second and extract one frame
[15:32:03 CET] <Elirips> but then I loose a lot of time as ffmpeg has to connect to the streams first etc
[15:32:12 CET] <temhawk> ffmpeg will grab a frame every ten seconds but the process will finish as fast as it can
[15:32:20 CET] <temhawk> it doesn't play the video in real-time
[15:34:20 CET] <temhawk> e.g. if you have a 1-minute video, you'll get an output.png file of the 51st frame after 2 seconds or so
[15:34:31 CET] <Elirips> temhawk: its a live-stream, that will never end
[15:35:24 CET] <Elirips> I need "someting" that connects to an rtsp stream, and stores one frame of the stream somewhere on a disk where apache can serve it then
[15:35:31 CET] <Elirips> The idea is that users see that as a "thumbnail"
[15:35:35 CET] <temhawk> Elirips: <mysource> is a fifo or socket?
[15:35:43 CET] <Elirips> rtsp
[15:36:18 CET] <Elirips> and the user can then "open" the live stream to view it, but like that he gets a more or less up-to-date thumbnail and has not to remember dozends of camera-ids
[15:36:59 CET] <Elirips> I could also let ffmpeg run constantly, using a pattern like %03.png, and then do some clever apache-rewriting to always serve the newest file (based on timestamp or so)
[15:37:10 CET] <Elirips> but I would prefer it if ffmpeg would just write to the same file over and over again
[15:38:38 CET] <temhawk> why not use ffplay and redirect output to a file?
[15:40:10 CET] <Elirips> hm, would this be possible with ffplay?
[15:41:27 CET] <temhawk> oh, uhm, NEVERMIND - NEVERMIND
[15:41:34 CET] <temhawk> that was dumb
[15:43:13 CET] <Elirips> Ah, I found this entry on stackoverflow: http://superuser.com/questions/592160/optimise-ffmpeg-live-stream-thumbnail
[15:43:17 CET] <Elirips> this is exactly what I need
[15:43:22 CET] <Elirips> more or less
[15:44:49 CET] <Elirips> .. this will get interesting once I have a lot of users accessing the thumbnails - who will win: apache trying to read the file or ffmpeg trying to write it..
[15:47:21 CET] <kepstin-laptop> might have to tweak the writer to do the "write to temp file and rename" dance if you want that to work correctly.
[15:47:28 CET] <Elirips> What is this 'updatefirst 1' param in the sample mentioned here: http://superuser.com/questions/592160/optimise-ffmpeg-live-stream-thumbnail
[15:48:05 CET] <Elirips> kepstin-laptop: yes, maybe it would still be better to make ffmpeg create - say 3 thumbnails, and apache shall always serve the last modified file
[15:48:42 CET] <kepstin-laptop> Elirips: dunno what that param is; the "image2" muxer options http://ffmpeg.org/ffmpeg-formats.html#image2-2 lists an "update" parameter that causes it to overwrite the same file
[15:50:05 CET] <Earthnail> Hey there. Running ffmpeg on my Mac tells me it needs about 6Mb of RAM. Running it on an Ubuntu server shows me it loads all kinds of shared libraries, and needs 300Mb of RAM. Any explanations?
[15:50:59 CET] <Elirips> Earthnail: yes, probably missinterpreting virtual and private size
[15:53:05 CET] <Earthnail> Elirips: possibly, yes.
[15:53:27 CET] <Earthnail> So I guess virtual memory is just not shown on the Mac?
[15:53:59 CET] <Earthnail> Elirips: private memory is about 15-17M on Ubuntu
[16:24:02 CET] <Elirips> Earthnail: to be honest, I never really got the difference between private, virtual, etc. I always have to look it up on stackoverflow.
[16:24:23 CET] <Elirips> But for example, on windows, the standard task-manager shows the one that is very low always (private I guess)
[16:25:07 CET] <Elirips> on windows, ffmpeg shows me to be using ~24 mb, while on debian 'top' shows a memory size of exactly 512MB for every ffmpeg process
[16:25:11 CET] <Earthnail> well, I don't get virtual memory (is that shared memory?), but I understand shared library, and I get private memory
[16:25:58 CET] <Elirips> Hm, could I tell ffmpeg to instead of "writing" the extracted frame, it shall pipe it to some socket?
[16:28:39 CET] <Elirips> Then I could write an apache mod that accepts the data from ffmpeg and always serves the last complete frame
[16:30:20 CET] <JohanW_Work> Elirips: When I'm extracting a frame I use -frames:v 1 -f image2 pipe:1
[16:30:58 CET] <JohanW_Work> IIRC the -f image2 is necessary to make the pipe work. :1 means stdout, if I don't misremember.
[16:32:45 CET] <JohanW_Work> Elirips: Here's the documentation for pipe: https://www.ffmpeg.org/ffmpeg-protocols.html#pipe
[16:32:48 CET] <Elirips> JohanW_Work: Great news, great hint
[16:33:04 CET] <Elirips> I guess thats the only "clean" solution to avoid locking problems.
[16:33:49 CET] <Elirips> I will pipe the frames to some process that also serves them over http
[16:43:49 CET] <Elirips> hm, could ffmpeg also send to a named pipe on windows?
[16:45:56 CET] <JohanW_Work> Elirips: I'm afraid I have no idea. I just managed to make it work on Linux, piping into php. You'll have to search for it.
[16:47:02 CET] <Elirips> JohanW_Work: hm, as on windows named pipes are just like "normal files", but being mounted on a special path like \\.\pipe\foo it should theoretically work... well, I'll try
[16:48:05 CET] <Elirips> hehe, it looks good
[16:48:52 CET] <Elirips> If I just try to make ffmpeg write to '\\.\pipe\foo.png' it fails with a 'Could not open file: \\.\pipe\foo.png'
[16:49:10 CET] <Elirips> thanks anyone for all the hints
[16:52:48 CET] <ramiro> Elirips, names pipes don't work on windows
[17:45:46 CET] <msocorcim> hi, I'm using the dvd-slideshow tool to convert a series of PNG images to a movie
[17:47:43 CET] <msocorcim> some obsolete arguments to ffmpeg and soxi needed to be fixed, ffmpeg was -loop_input -> -loop 1, soxi was just changing the order of some arguments from something that never should have worked to the currently supported syntax
[17:48:20 CET] <msocorcim> the process uses ffmpeg to generate some intermediate PPM images
[17:49:14 CET] <msocorcim> the source PMG images have nice crisp text, where the PPM images have blurry artifacted text
[22:18:03 CET] <sergio-br2> Hi, how can I convert these negative errors? I got -1330794744 (0x4F5250F8)
[22:18:10 CET] <sergio-br2> No URL Protocols are registered. Missing call to av_register_all()?
[22:19:18 CET] <Qazerowl> What VBR audio quality is closest to 192K? I read that VBR 5 is roughly 128K, so is 192K about the same as VBR 4? Or 3?
[22:20:15 CET] <BtbN> 192k means nothing in terms of quality, it's just a bitrate.
[22:20:28 CET] <BtbN> Depends entirely on the codec what quality that is with cbr.
[22:21:15 CET] <Qazerowl> m4a/acc?
[22:21:59 CET] <BtbN> https://trac.ffmpeg.org/wiki/Encode/AAC#fdk_vbr
[22:22:07 CET] <BtbN> again, only applies to that encoder.
[22:22:29 CET] <BtbN> 192k is kind of overkill for aac with a decent encoder.
[22:22:45 CET] <BtbN> 128k cbr is already basicaly perfect in terms of noticable artifacts.
[22:29:30 CET] <Qazerowl> Alright, thank you.
[22:39:20 CET] <MadTBone> can anyone recommend a high quality, low complexity (as in fast) intraframe format for files of arbitrary resolution?  I'm currently using DNxHD (quite successfully) for 1080p files, but I need to support odd resolutions like 1920x540 and 1280x360.  Space is a consideration. I can go as high as ~200Mb/s for 1080, but would prefer to stay down around 100Mb/s.  It must be able to do at least 2x realtime framerate (at 1920x540)  on a modest modern CPU (Has
[22:39:20 CET] <MadTBone> well i5 w/ 8 logical cores).
[22:41:29 CET] <ChocolateArmpits> MadTBone: Prores, while the white paper outlines resolution limits, the encoders available are flexible
[22:43:35 CET] <MadTBone> ChocolateArmpits: I've tinkered with it in the past, and found it to be much slower than DNx for 1080p.  That was about 2 years ago.  Has the speed improved in the ffmpeg implementation?
[22:44:16 CET] <ChocolateArmpits> MadTBone: if you tried using prores_ks (kostyia) then it was most likely slow because of low default qscale
[22:44:53 CET] <BtbN> h264 isn't an option?
[22:45:07 CET] <MadTBone> ChocolateArmpits: ahh, as in extra effort to eek out 3 or 4 q at the same bitrate?
[22:45:15 CET] <ChocolateArmpits> BtbN:  It is if you set gop size to 1
[22:45:39 CET] <BtbN> I know, that's why i'm asking.
[22:46:18 CET] <ChocolateArmpits> MadTBone: low qscale forces bitrate conservation at expense of processing speed
[22:46:35 CET] <ChocolateArmpits> Read this, there's a section "how to make it encode faster" http://codecs.multimedia.cx/?p=388
[22:47:08 CET] <MadTBone> BtbN:  264 intra might be an option for me.  but the speed is a hard requirement.  this is for continuous live ingest from multiple sources, and again, like Prores, in the past I've found it too slow.
[22:47:24 CET] <ChocolateArmpits> There's another encoder prores_aw (anatolyi), and while that's much faster by default, maybe perhaps due to multithreading, it's more feature limited, no support for interlaced or 4444 profile, also lower quality
[22:47:31 CET] <BtbN> Speed as in CPU usage, or faster-than-realtime fps?
[22:47:48 CET] <ChocolateArmpits> default qscale is at least 2 times slower than realtime :D
[22:47:50 CET] <MadTBone> faster than realtime
[22:48:00 CET] <BtbN> You could go for some hw decoder.
[22:48:09 CET] <MadTBone> $
[22:48:15 CET] <BtbN> *encoder
[22:48:22 CET] <BtbN> nvenc does ~800 FPS in lossless mode at 50-100 Mbps
[22:48:26 CET] <ChocolateArmpits> MadTBone: Why not intra mpeg2  ?
[22:48:40 CET] <MadTBone> ChocolateArmpits:  is that even a thing?!
[22:48:48 CET] <ChocolateArmpits> As long as you can set the gop size it is
[22:48:54 CET] <BtbN> (lossless mode isn't implemented in the ffmpeg encoder though)
[22:49:37 CET] <MadTBone> nvenc = nvidia?  no gpu here, nor is there the ability to add one
[22:51:40 CET] <ChocolateArmpits> If you have an Intel processor with intel graphics quick sync should be available to do hardware h264 encode
[22:51:45 CET] <MadTBone> what about support for quick sync
[22:51:48 CET] <MadTBone> ahh...
[22:51:55 CET] <MadTBone> beat me to the punch
[22:52:05 CET] <BtbN> There is no QuickSync on linux.
[22:52:15 CET] <BtbN> There is libva, but hols shit that library...
[22:52:17 CET] <MadTBone> there is, but in ffmpeg?
[22:52:17 CET] <BtbN> *y
[22:52:26 CET] <BtbN> Nope, no QSV on linux.
[22:54:31 CET] <MadTBone> I assume the same is true for bsd?
[22:54:47 CET] <BtbN> On bsd there propably isn't even libva.
[00:00:00 CET] --- Tue Mar 24 2015


More information about the Ffmpeg-devel-irc mailing list