[Ffmpeg-devel-irc] ffmpeg.log.20150326

burek burek021 at gmail.com
Fri Mar 27 02:05:01 CET 2015


[00:00:06 CET] <DragonsLordKERNE> seasc, it's a AVInaptic report
[00:00:32 CET] <entrailz> I believe I have that in place as it would not present a different error, i will run it again now and pastebin the error if i remove that.
[00:00:37 CET] <DragonsLordKERNE> seasc, you can also receive a warning on VBR if you open the resulted output with VirtualDubMOD
[00:01:26 CET] <seasc> DragonsLordKERNE, thank you, but its windows.. :(
[00:02:57 CET] <seasc> entrailz, i was just asking, i have no idea what it does ^^
[00:03:22 CET] <entrailz> seasc, its something to do with the bitstream, i believe i had it in place as it was throwing an error before
[00:03:43 CET] <DragonsLordKERNE> seasc, AVInaptic is also available on Linux
[00:04:02 CET] <DragonsLordKERNE> I am a *nix man at work ... obliged to use Windows at home :-/
[00:09:29 CET] <seasc> DragonsLordKERNE, just found it, sadly there is no package in rpmfusion, so its not worth the effort, just wanted to have a look at the output of one of my files to compare. But comiling is too much trouble for this :p
[00:12:09 CET] <Neon> seasc, Thanks, btw, I'm compiling now :)
[00:12:40 CET] <DragonsLordKERNE> seasc, I understand ;)
[00:12:58 CET] <entrailz> seasc, iive, thanks for help but seems im not destined to resolve this today. thanks anyway!
[00:13:04 CET] <seasc> DragonsLordKERNE, if you use ffmpeg at work too, you might like my bash script project : https://github.com/sri-arjuna/vhs/blob/master/vhs.sh
[00:14:26 CET] <seasc> Neon, welcome :) btw, i guess you might like my above script as well ;)
[00:15:39 CET] <seasc> there's a manpage and bash_completion as well in the git repo :)
[00:16:29 CET] <seasc> though its only for 'general use' nothing specialiced
[00:16:57 CET] <DragonsLordKERNE> thank you for your suggestion, seasc
[00:17:43 CET] <DragonsLordKERNE> Well, time to go to bed guys ... I really appreciated your help ... Good night! :)
[00:18:37 CET] <seasc> DragonsLordKERNE, good night :)
[00:19:17 CET] <seasc> and sorry about cbr/vbr ..
[00:19:30 CET] <seasc> oh wait.. isnt there a cbr 23 setting or so? or does that only apply to video?
[00:20:09 CET] <DragonsLordKERNE> I don't really know ... what I know is that using AVIdemux the result it's a "real" CBR
[00:20:38 CET] <DragonsLordKERNE> the result by FFMPEG seems to be good but not well "attributed"
[00:21:07 CET] <DragonsLordKERNE> now I really have to leave sorry ... thank you very much again ... talk to you next time ;)
[00:21:19 CET] <DragonsLordKERNE> Click!
[00:21:25 CET] <seasc> gn8
[00:21:42 CET] <selsta> can i use lavf to use a byte array insted of a file path?
[00:22:31 CET] <selsta> when using something like avformat_open_input
[01:01:02 CET] <IlIlIlIlIlIlI> re topic: ffmpeg is in debian unstable already ^^
[01:01:13 CET] <IlIlIlIlIlIlI> what does "Non-monotonous DTS in output stream"  mean?
[01:01:40 CET] <IlIlIlIlIlIlI> particularly as only manifests when using '-f tee' format
[03:25:54 CET] <Anaphaxeton> hello. i have a question
[03:26:24 CET] <Anaphaxeton> is there a trick to ruen an srt to a dvdsub?
[03:26:31 CET] <Anaphaxeton> no video embedding
[03:26:53 CET] <Anaphaxeton> s/ruen/turn
[03:27:03 CET] <maqr> how can i tell what "level" this h.264 stream is encoded with? it says "Video: h264 (High) (avc1 / 0x31637661), yuv420p(tv, smpte170m/bt709/bt709), 640x480, 5008 kb/s, 29.88 fps, 30 tbr, 600 tbn, 1200 tbc (default)", but i was expecting to see something like "Level 3.1" after "High"
[03:32:19 CET] <lordkrondor>  use the -show_streams option, it will show you the level=
[03:32:40 CET] <lordkrondor> for ffprobe that is
[03:39:30 CET] <maqr> thanks lordkrondor
[03:42:01 CET] <maqr> lordkrondor: "level=30" must mean 3.)?
[03:42:03 CET] <maqr> 3.0?
[03:42:14 CET] <lordkrondor> yeah from my experience
[03:42:20 CET] <maqr> ok, cool
[03:47:04 CET] <Neobenedict> [02:37:11] <+Neobenedict> hi all, I run this ffmpeg -i input.mkv -ss 1:07:34 -t 00:00:40 -c:v libx264 -c:a copy out.mkv and get ffmpeg freezing, any idea why
[03:47:04 CET] <Neobenedict> [02:37:26] <+Neobenedict> stuck here Press [q] to stop, [?] for help
[03:47:08 CET] <Neobenedict> [02:37:26] <+Neobenedict> frame=    0 fps=0.0 q=0.0 size=       1kB time=00:00:00.00 bitrate=N/A
[03:47:21 CET] <Neobenedict> [02:40:06] <+Neobenedict> oh now we got this
[03:47:22 CET] <Neobenedict> [02:40:08] <+Neobenedict> Reference 6 >= 50.0 q=0.0 size=       1kB time=00:00:00.00 bitrate=N/A
[03:47:22 CET] <Neobenedict> [02:40:08] <+Neobenedict> [h264 @ 000000001128ac40] error while decoding MB 77 22, bytestream (98391)
[03:47:22 CET] <Neobenedict> [02:40:08] <+Neobenedict> [h264 @ 000000001128ac40] concealing 3332 DC, 3332 AC, 3332 MV errors in P frame
[03:47:30 CET] <Neobenedict> this channel might be better
[03:58:58 CET] <Neobenedict> klaxa
[03:58:59 CET] <Neobenedict> http://privatepaste.com/96136104d1
[03:59:54 CET] <klaxa> Neobenedict: that shows no error at all
[04:00:07 CET] <Neobenedict> but it shows it stuck
[04:00:11 CET] <Neobenedict> just sits there
[04:00:26 CET] <Neobenedict> okay it's doing something now
[04:00:38 CET] <Neobenedict> weird considering i had it sat for ages
[04:01:33 CET] <klaxa> it will sit there for quite a long time, it's decoding the frames up to 1:07:34
[04:01:43 CET] <klaxa> you can move the -ss option before the -i option
[05:12:58 CET] <Guest24492> Hi, I'm trying to chain together multiple -filter_complex in the one ffmpeg command. Can you advise if this is possible? ffmpeg -y -i /mnt/audio/Programs/23/23051/A23051/PM/A23051-01.wav  -filter_complex 'compand=0.005:0.3:-13.5/-13.5|0/-12.5:1:0:-90:0' -filter_complex 'compand=0.002:.15:-12.5/-12.5|0/-11.5:1:0:-90:0' /media/local_temp/1.wav
[05:20:12 CET] <klaxa> you can use -filter_complex only once, you can cobine the two arguments into one separated by a semicolon (i think? might be comma)
[05:25:15 CET] <Guest24492> Thanks. ; didn't work. , does. Great!
[07:41:40 CET] <maqr> is there some standard way to rate visual quality of encoded video?
[07:42:02 CET] <maqr> i think it's mostly subjective, but there might be something that i'm missing
[07:43:27 CET] <vivan> objective metrics like psnr and ssim (but they are not as good from subjective point of view)
[07:49:18 CET] <maqr> vivan: interesting, thanks, i googled about that and landed here http://en.wikipedia.org/wiki/Video_quality
[07:50:03 CET] <maqr> it looks like PSNR is what i was looking for on the objective side, and MOS is the subjective version
[07:51:48 CET] <maqr> http://en.wikipedia.org/wiki/PEVQ this is also a thing
[07:58:05 CET] <maqr> apparently there's a thing called the Video Quality Experts Group
[09:07:45 CET] <JEEBsv> maqr: there's always issues with metrics, like PSNR always favouring a blurred image over a sharper one with some artifacts
[09:08:29 CET] <JEEBsv> SSIM seems to be a bit better in some things, but it's definitely not an answer to all the questions
[09:26:21 CET] <termos> what's the difference of using "," and ";" in my filter graph string?
[09:41:49 CET] <_pekman_> hello everyone, anyone there?
[09:42:59 CET] <_pekman_> I'm trying conatenar a * .wav and * mp4 video using mmcat script, but I get an error
[09:43:43 CET] <_pekman_> see http://imgur.com/4Tm3855
[09:44:09 CET] <_pekman_> /tmp/mcs_v_all: Operation not permitted
[10:03:58 CET] <relaxed> _pekman_: it's probably a permission problem.
[10:08:16 CET] <_pekman_> is not a permission issue, I tried to accomplish the task as root and got the same error
[10:20:28 CET] <relaxed> hmm, maybe you can't have fifos in /tmp
[10:21:13 CET] <relaxed> change /tmp in the script to point to your home dir or $HOME/tmp
[10:29:28 CET] <_pekman_> even making the change does not work
[10:30:09 CET] <_pekman_> there is a simpler way to concatenate * .wav and * .mp4?
[10:30:37 CET] <_pekman_> I'm making a script to record the screen
[10:31:04 CET] <relaxed> concat wav with the audio in the .mp4?
[10:31:37 CET] <_pekman_> as not use pulse, had to do different things in order to capture my audio
[10:32:05 CET] <relaxed> so there's no audio in the .mp4 and you want to add it via the wav?
[10:32:20 CET] <_pekman_> * wav audio and video * .mp4
[10:32:36 CET] <_pekman_> yes
[10:33:09 CET] <_pekman_> ffmpeg -f alsa -i default:CARD=MID /tmp/out-`date +%d-%m-%Y_%H:%M:%S`.wav -f x11grab -r 30 -s `xdpyinfo | grep 'dimensions:'|awk '{print $2}'` -i :0.0  -an -vcodec libx264 -preset ultrafast -qp 0 -threads 0 /tmp/screencast-`date +%d-%m-%Y_%H:%M:%S`.mp4
[10:33:23 CET] <_pekman_> i using this command in shell script
[10:33:44 CET] <relaxed> ffmpeg -i video.mp4 -i audio.wav -map 0 -map 1 -c:v copy -c:a aac -strict -2 output.mp4
[10:33:46 CET] <_pekman_> so far everything works perfectly
[10:34:01 CET] <_pekman_> ooohhhh
[10:34:14 CET] <_pekman_> I will try it
[10:37:01 CET] <relaxed> concat is where you add one stream to another, making it longer.
[10:38:27 CET] <relaxed> muxing it what you needed
[10:38:46 CET] <_pekman_> what do you suggest?
[10:39:25 CET] <relaxed> the command I just gave you muxes both audio and video into output.mp4
[10:41:24 CET] <_pekman_> well! I have to tell you that did not know this command
[10:42:41 CET] <relaxed> the command you use includes audio, so where does the wav come from?
[10:42:59 CET] <relaxed> oh, I see
[10:43:45 CET] <relaxed> ffmpeg -f alsa -i default:CARD=MID -f x11grab -r 30 -s `xdpyinfo | grep 'dimensions:'|awk '{print $2}'` -i :0.0  -an -vcodec libx264 -preset ultrafast -qp 0 -threads 0 output.mp4
[10:44:15 CET] <relaxed> ^^ that would encode audio and video into output.mp4
[10:54:01 CET] <un1versal> morning, Would like to knw if thers circumstances where .wav files get identified as having a wav audio stream by ffmpeg and not pcm_s16le or pcm_s24le etc
[10:56:09 CET] <_pekman_>  relaxed the command to encode audio and video in mp4 does not work. Apparently I have to record wav mp4 and then use muxes.
[11:05:48 CET] <relaxed> _pekman_: in what way does it not work?
[11:06:22 CET] <_pekman_> the audio is muted
[11:06:36 CET] <_pekman_> the video is muted
[12:33:54 CET] <JamJams> Is there a way for ffmpeg to keep field repeats? When it decodes it ignores them, it's highly annoying.
[13:16:51 CET] <JamJams> How do I handle RFF pull-down with decimate in -vf?
[13:30:33 CET] <lordkrondor> use fieldmatch followed by decimate
[13:30:53 CET] <lordkrondor> and I also think you need to set -r 24000/1001
[13:31:03 CET] <JamJams> That deletes good frames though
[13:31:13 CET] <JamJams> Since it has repeat fields the decimate pattern is wacked
[13:31:19 CET] <lordkrondor> how about just use pullup
[13:31:21 CET] <JamJams> ffmpeg ignores repeat fields
[13:31:28 CET] <JamJams> That also results in the same issue
[13:31:35 CET] <lordkrondor> sorry, don't know then
[13:33:03 CET] <JamJams> Seems like a pretty huge flaw
[13:55:07 CET] <JEEBsv> JamJams: there is a field match filter and a decimation filter
[13:55:31 CET] <JamJams> ffmpeg ignores field repeats
[13:55:49 CET] <JamJams> There is no code to handle it at all if you look at the mpeg2dec.c
[13:55:58 CET] <JEEBsv> oh right, RFF stuff
[13:56:01 CET] <JamJams> So it doesn't even know the frames are there
[13:56:16 CET] <JamJams> So if you apply a decimate filter over the top of it, the pattern is screwed
[13:56:23 CET] <JEEBsv> yeah
[13:56:30 CET] <JamJams> I don't know how to fix it :/
[13:56:47 CET] <JEEBsv> I just remember some lavc-using things somehow hacking that in
[13:57:02 CET] <JEEBsv> but that is probably manual parsing of the flags
[13:57:59 CET] <JamJams> I've set my source to 24000/1001 and just yadif'd it, there's still dupes in there
[13:58:52 CET] <JamJams> Why doesn't ffmpeg just dupe the frame though
[13:59:16 CET] <JamJams> Somebody commited one change a few years ago and got rid of all the field repeating code
[13:59:30 CET] <JamJams> Before that it had something that would insert the frame again or a null packet
[14:08:36 CET] <JamJams> JEEBsv do you think I could work my way around it by applying yadif then attempting to remove dupe frames with decimate somehow?
[14:39:46 CET] <iive> I think that the decoder exports the flags. There was a mplayer filter that creates telecine from them...
[14:40:26 CET] <iive> but i'm not sure if that filter was ported, before removal.
[14:42:12 CET] <JamJams> I don't really care about the flag or recreating the pull down
[14:42:17 CET] <JamJams> I just want to be able to get rid of the dupe frames
[14:42:43 CET] <JamJams> Without the proper frame number sequence decimate can't do it's job
[14:44:07 CET] <JamJams> iive if the flag is exported how can I access it? Is it stored in AVFrame?
[14:51:23 CET] <iive> there must be mpdecimate filter
[14:51:30 CET] <iive> that looks at the content of the frames
[14:51:45 CET] <JamJams> There is, but it deletes too much
[14:52:08 CET] <He4dShOt> hello
[14:52:14 CET] <iive> it should have some thresholds to tweak
[14:52:37 CET] <JamJams> Yeah, but it wont ever be 100%, not respecting the repeat fields should be classed as a bug
[14:52:47 CET] <He4dShOt> can I find ffmpeg already compiled for raspberry pi? I'm in a hurry and don't really have the time to compile it
[14:52:50 CET] <JamJams> You can't fix ITVC if the frame seq is lost
[14:53:40 CET] <iive> hum?
[14:55:38 CET] <iive> the decimate filter is smarter than fps that simply drops frames in sequence, it should instead drop the most similar frames first.
[14:57:07 CET] <JamJams> The sequence is lost though, if you have a RFF pull-down then apply decimate over it. It's going to drop valid frames as well as the dupes
[14:57:22 CET] <JamJams> Because ffmpeg ignores the repeat_field flag
[14:57:35 CET] <iive> I don't understand what do you mean by "the sequence"
[14:57:52 CET] <iive> do you want to reconstruct the original frames or not?
[14:59:43 CET] <JamJams> But it's RFF, ffmpeg ignores the repeat_field flag in the mpeg12enc.c so it's impossible to restore it to the original
[14:59:52 CET] <iive> why?
[15:00:06 CET] <iive> the decoder output all frames as they have been encoded
[15:00:21 CET] <JamJams> The decoder ignores the repeated fields
[15:00:37 CET] <iive> because these fields are not encoded
[15:00:51 CET] <iive> it just means that the frame should stay a half period longer
[15:01:00 CET] <JamJams> Correct they're just duplicate
[15:01:16 CET] <iive> but, does the decoder output duplicate?
[15:01:21 CET] <iive> imho, it should not.
[15:01:25 CET] <JamJams> No it doesn't and that's the problem
[15:01:31 CET] <iive> why?
[15:01:55 CET] <iive> you have 30 fps, the decoder output 24 and you have decimate that is set to 24...
[15:02:02 CET] <JamJams> Because there's still duplicates in the source
[15:02:29 CET] <JamJams> Sometimes the repeat field flag isn't set true by the creator of the .ts file
[15:02:37 CET] <JamJams> Sometimes they actually just duplicate the frame
[15:02:47 CET] <iive> you mean, hard telecine
[15:02:50 CET] <JamJams> Yes
[15:03:04 CET] <JamJams> It's a mix of both hard telecine and rff
[15:03:45 CET] <JamJams> And because ffmpeg doesn't respect the repeated fields flag, you can't apply decimate over the top of it to remove the hard telecine and the soft telecine
[15:04:25 CET] <JamJams> The sequence is lost so you end up destroying valid frames if you try to remove the hard telecine mixed in with the repeats that ffmpeg doesn't see
[15:05:08 CET] <iive> as I said, decimate should remove frames that are most similar first.
[15:05:39 CET] <iive> also... aren't some of the special ivtc filter better suited for handling such cases?
[15:06:24 CET] <JamJams> I could use .avs and it respects the pull down so I can just apply decimate but I want to do it all with ffmpeg
[15:06:47 CET] <JamJams> I just think it's odd there's no filter or flag to make it respect the pull down and show the repeats
[15:07:09 CET] <iive> well... the ffmpeg decimate should behave the same as avisynth one.
[15:07:18 CET] <iive> or at least that was the idea...
[15:07:58 CET] <JamJams> avisynth respects the pull down so decimate can do it's job, ffmpeg doesn't respect the pull down so decimate can't
[15:08:23 CET] <iive> this doesn't make sense
[15:08:57 CET] <JamJams> Ok so avisynth will do this 1 2 3 4 5 FIELDREPEAT-5 6 7 8 9 10
[15:09:18 CET] <JamJams> ffmpeg does this 1 2 3 4 5 FIELDREPEAT-5-IGNORE 6 7 8 9 10
[15:10:51 CET] <JamJams> So when you throw decimate over ffmpeg it's already removed 98% of the frames by ignoring the pull down
[15:10:57 CET] <JamJams> So decimate doesn't work
[15:11:00 CET] <JamJams> That's the problem
[15:11:05 CET] <iive> decimate filter needs constant fps input... so i guess this is the problem
[15:11:06 CET] <JamJams> And I can't see a way around it
[15:11:35 CET] <JamJams> There should be a filter or something or a flag in the decoder
[15:11:40 CET] <JamJams> So you can add the repeats
[15:12:25 CET] <iive> try the 'pullup' filter
[15:13:23 CET] <JamJams> Any extra parameter I should try or just -vf pullup
[15:14:07 CET] <iive> try it without any, see if there are places where it does (not) work, tweak for them.
[15:18:04 CET] <iive> there is 'telecine' filter, but I don't think it handles rff flags :(
[15:18:26 CET] <JamJams> Nothing does :/
[15:19:00 CET] <JamJams> mpeg12dec.c line 1140
[15:19:13 CET] <JamJams> You can't even get the flag in AVFrame or AVPacket to fix it yourself in a filter
[15:19:18 CET] <JamJams> It's just always set to 0
[15:19:43 CET] <JamJams> I'm not even sure why whoeever wrote it wastes the memory used to store the int if it's always just 0
[15:23:25 CET] <iive> it's repeat_first_field
[15:25:51 CET] <JamJams> So what's repeat_field then :/
[15:26:21 CET] <iive> no idea
[15:26:46 CET] <baran> whats the difference between yuv420p and yuv420p(tv) ?
[15:28:03 CET] <iive> baran: one of them have dead zones, aka all values from 0 to 16 are considered black, and all from 240 to 255 are considered while.
[15:28:40 CET] <iive> so the range is smaller 16-240. Not sure if these values are correct or which one is the one with/out dead zones.
[15:29:43 CET] <JamJams> Hmm well I can't access repeat_first_field outside the class I don't think
[15:29:59 CET] <loki_> hello.
[15:30:14 CET] <JamJams> Guess I can just edit this to reinsert the last frame
[15:30:14 CET] <loki_> if x11grab slow method to capture screen?
[15:30:40 CET] <loki_> i want do 60 fps capture of my gameplay
[15:30:59 CET] <loki_> i made nvenc version of ffmpeg but can't get even 30 fps
[15:31:52 CET] <BtbN> How are you trying to capture, and what platform?
[15:33:04 CET] <JamJams> @iive - pullup doesn't work, it leaves interlaced frames so I added yadif=1:1 now it leaves dupes
[15:33:04 CET] <baran> iive: is there a way to tell one format from another in ffmpeg? To my knowlage both yuv420p and yuv420p(tv) are represented as "AV_PIX_FMT_YUV420P"
[15:33:16 CET] <baran> or am i mistaken?
[15:36:31 CET] <BtbN> loki_, even on the slowest setting (llhq with 2pass), 60 fps with 720p are still possible. So i'd assume your capture isn't fast enough.
[15:38:16 CET] <loki_> BtbN: ubuntu 14.04 ./ffmpeg -f x11grab -r 60 -s 1920x1080 -i :0.0 -vcodec libnvenc -b:v 3050k /media/loki/Files/test1.avi
[15:38:54 CET] <loki_> intresting things that that old not official nvenv repo is faster thatn official.
[15:39:01 CET] <BtbN> x11grab can have a hard time getting 60 fps.
[15:39:47 CET] <BtbN> How many FPS do you get if you just copy the raw frames to /dev/null?
[15:40:06 CET] <loki_> how i can do this?
[15:40:15 CET] <BtbN> -c copy instead of some encoder
[15:41:25 CET] <loki_> is this right? ./ffmpeg -f x11grab -r 60 -s 1920x1080 -i :0.0 -c copy /dev/null
[15:41:37 CET] <loki_> it said it can't found suitable format
[15:41:46 CET] <BtbN> ffmpeg -f x11grab -r 60 -s 1920x1200 -i :0 -c copy -f rawvideo /dev/null
[15:41:47 CET] <BtbN> works for me
[15:45:06 CET] <loki_> BtbN: and how much you have? http://pastebin.com/f2e3MyER
[15:45:14 CET] <loki_> i can't get it working
[15:45:26 CET] <BtbN> I get 55 fps here.
[15:46:21 CET] <loki_> what cfg and os you have?
[15:46:32 CET] <loki_> and what is my problem with that command?
[15:46:56 CET] <BtbN> Is that the ffmpeg you also use nvenc with? Cause 2.4 doesn't have nvenc.
[15:47:38 CET] <BtbN> I'm using 2.6.1
[15:48:38 CET] <loki_> no it is system ffmpeg
[15:49:58 CET] <loki_> i have 75 fps(
[15:50:07 CET] <loki_> so nvenc is bottleneck?
[15:51:24 CET] <loki_> and ffmpeg eating around 40% of cpu
[15:51:37 CET] <loki_> with nvenc i see 60%
[15:51:59 CET] <hurn> hello, im currently making a program using ffmpeg. i was wondering is there a way i can keep my portion of the code closed source?
[15:52:52 CET] <kepstin-laptop> hurn: use a build of ffmpeg that has an LGPL license, and read through http://ffmpeg.org/legal.html
[15:54:41 CET] <loki_> x11grab is really fast thing..
[15:54:53 CET] <loki_> it even can do 80 fps
[15:54:57 CET] <loki_> i am amused
[15:55:08 CET] <loki_> so maybe i can tweak nvenc?
[15:55:44 CET] <loki_> or it is just in early stage and i need to wait?
[15:56:25 CET] <BtbN> Try other presets, but even the slowest one should be easily able to do 60fps
[15:56:37 CET] <BtbN> Try -preset hq and -preset hp
[15:58:35 CET] <hurn> kepstin-laptop: so as long as i compile it as a dll under lgpl, ans say in the program about box. that i use ffmepg, ill be good? also i read that i have to include the source code i used for ffmpeg, but is that only if i modify ffmpeg? or no matter what i have to include it?
[15:59:02 CET] <kepstin-laptop> hum: if you're distributing a binary build of ffmpeg, you should always be distributing the corresponding source code as well
[15:59:16 CET] <kepstin-laptop> hurn: read the legal page, it says that right on it...
[15:59:41 CET] <hurn> i did just making sure i read it right...
[16:03:19 CET] <BtbN> If you built it from unmodified official sources, a short text file which says so and some info about which commit hash you build should be enough. I'm not a legal expert though.
[16:03:57 CET] <loki_> BtbN: i can't get nvenc some much more that 25 fps
[16:04:01 CET] <kepstin-laptop> it's certainly never wrong to distribute the full source yourself, of course.
[16:04:07 CET] <loki_> it really always equal to 25 fps
[16:04:09 CET] <BtbN> loki_, what GPU, what driver version?
[16:04:10 CET] <loki_> it is strange
[16:04:29 CET] <loki_> 346.47 660 gtx
[16:04:38 CET] <BtbN> built with nvenc api 5?
[16:04:50 CET] <loki_> i tried both
[16:04:57 CET] <loki_> official with api 5
[16:05:05 CET] <loki_> not official with api 4
[16:05:15 CET] <loki_> and not official is little faster
[16:05:47 CET] <loki_> if i change quality nothing happens
[16:05:53 CET] <loki_> always 25fps
[16:06:52 CET] <t4nk518> Hello, can I ask someone's help to work out an FFMPEG cli command?
[16:07:09 CET] <loki_> i know that it can do more i tried nvenc on this system under windows. And i recieved something near 60 fps
[16:09:23 CET] <BtbN> loki_, my gt630 does 77 fps on the slowest settings.
[16:09:33 CET] <loki_> brrr
[16:09:42 CET] <BtbN> 210 fps on the fastest
[16:09:47 CET] <loki_> where can i start again?
[16:09:59 CET] <BtbN> Are you running a game on the GPU?
[16:10:13 CET] <loki_> i am trying to capture just desktop
[16:10:46 CET] <BtbN> https://bpaste.net/show/28cf7645855e
[16:11:49 CET] <BtbN> Maybe your system is already overwhelmed by the rgb->yuv conversion?
[16:12:36 CET] <loki_> BtbN: what it mean?
[16:13:07 CET] <loki_> i have cpu=native in my configuration options
[16:13:15 CET] <loki_> can be that a problem?
[16:14:51 CET] <BtbN> What CPU is it?
[16:15:22 CET] <loki_> dunno
[16:15:43 CET] <loki_> wait lets start from build, plz
[16:15:57 CET] <loki_> BtbN: under what operation system you build it?
[16:16:08 CET] <BtbN> Try -c:v rawvideo -pix_fmt nv12 -f rawvideo -y /dev/null
[16:16:24 CET] <BtbN> I'm on Gentoo, using gcc 4.9.2
[16:16:40 CET] <BtbN> ffmpeg built from latest master
[16:16:57 CET] <loki_> how you build ffmpeg? from where you take all sources for example nvenc sdk?
[16:17:07 CET] <loki_> gentoo have it in portage or what?
[16:17:11 CET] <BtbN> There is only exactly one source for the nvenc sdk.
[16:17:21 CET] <loki_> nvidia site?
[16:17:54 CET] <BtbN> ffmpeg -f x11grab -r 60 -s 1920x1080 -i :0 -c:v rawvideo -pix_fmt nv12 -f rawvideo -y /dev/null
[16:17:59 CET] <BtbN> How many fps does this give you?
[16:18:50 CET] <loki_> lol
[16:18:51 CET] <loki_> 25
[16:19:08 CET] <BtbN> So your CPU is too slow to handle rgb to yuv fast enough.
[16:19:36 CET] <BtbN> Or you broke acceleration
[16:20:06 CET] <loki_> i see that now ffmpeg eat more that when i use nvenc
[16:20:20 CET] <kepstin-laptop> rgb to yuv conversion seems like something which it would be ideal to do on a wide vector processor attached over pci-e rather than the main cpu, tho. I'm surprised nvenc doesn't accept rgb input.
[16:20:24 CET] <loki_> maybe i need use cuda for this?
[16:20:38 CET] <BtbN> kepstin-laptop, it accepts only nv12
[16:20:44 CET] <BtbN> and yuv444p in lossless mode
[16:21:00 CET] <BtbN> (Or Direct3D surfaces)
[16:21:26 CET] <loki_> why that on windows all is fast enougth?
[16:21:34 CET] <loki_> maybe i can run another tests?
[16:21:52 CET] <loki_> i see that under native system ffmpeg i recieve 23 fps, under mine 25 fps
[16:22:16 CET] <BtbN> No idea? I'd guess you somehow broke the asm optimizations.
[16:23:44 CET] <loki_> avconv gives 26
[16:24:06 CET] <BtbN> avconv is not ffmpeg.
[16:25:51 CET] <loki_> what cpu you have?
[16:26:20 CET] <BtbN> Intel(R) Core(TM) i3-3220 CPU @ 3.30GHz
[16:27:18 CET] <loki_> i have old phenom
[16:27:23 CET] <loki_> this can be a problem?
[16:27:41 CET] <loki_> AMD Phenom II X4 925
[16:28:51 CET] <BtbN> Propably, those things don't support a lot of stuff, so most hand written asm code won't be used there.
[16:35:21 CET] <BtbN> The bgr to nv12 conversion is doing 450 fps for me, btw.
[16:38:24 CET] <loki_> looks like i need upgrad lol.
[16:46:50 CET] <loki_> BtbN: i can't believe that there can be so much big difference.
[16:46:54 CET] <loki_> it is icredible.
[17:01:39 CET] <Genscher> is it a known issue, that ffprobe seems to report inacurate millisecond durations for mpeg-2 videos? (premiere + other app report e.g. XX.24 and ffprobe reports XX.296)
[17:02:33 CET] <Genscher> (which results in a black / empty frame when trusting ffprobe)
[17:14:37 CET] <micechal_> Is it possible to scale an image only if is larger than the destination size¿
[17:16:47 CET] <saste> micechal_, sure, check the scale filter docs
[17:17:28 CET] <micechal_> Im doing so right now
[17:30:25 CET] <micechal_> Seems I'll have to use the min() function, right?
[17:38:32 CET] <ausjke> i'm using avi for mjpeg streaming where video and audio are interleaved, is it possible to have audio track in avi for this use case? or some other format/container that can stream mjpeg with audio track?
[17:39:14 CET] <ausjke> due to the nature of AVI the mjpeg and audio are now interleaved, so I can't really extract the audio track in players such as vlc
[17:41:05 CET] <NY152> hello :)
[17:44:00 CET] <NY152> I have problems with ffmpeg (Windows). When I want to make an x264 encoding (libx264) ffmpeg puts lavc as encoder :/ libx264 is listed yet.
[17:50:38 CET] <loki_> BtbN: how can i understand that tech ffmpeg use for rgb to nv12 convertation?
[17:54:45 CET] <NY152> nobody ? :/
[18:00:54 CET] <loki_> BtbN: am i right that where all stuff around rgb to nv12 going is libswscale/x86/input.asm
[18:00:58 CET] <loki_> ?
[18:17:35 CET] <BtbN> loki_, it's all somewhere in libswscale, yes
[18:37:19 CET] <loki_> BtbN: looks i have no sse3 and it is a problem.
[18:37:29 CET] <loki_> looks like *
[18:37:39 CET] <BtbN> Realy, it doesn't even have sse3?
[18:39:21 CET] <loki_> oh no it must have ssse3
[18:39:28 CET] <loki_> not sse3
[18:39:35 CET] <loki_> and i have not.
[18:58:13 CET] <spectromas> hello, I want to convert the audio in a video file to mp3 V0 but keep the video and subs exactly how they are already. I want to check here before doing it because it's a large file and will take a long time I expect. This is what I was intending to do: ffmpeg -i /path/to/file -vcodec copy -scodec copy -acodec libmp3lame -qscale 0 /path/to/output/file
[19:02:33 CET] <spectromas> oh sorry, http://pastebin.com/BnivV51v
[19:20:37 CET] <spectromas> anyone? I'm guessing there is no need for profiles or presets, that this would be the quickest way?
[19:28:26 CET] <loki_> BtbN: how i can read this output pipe in another copy of ffmpeg? ffmpeg -f x11grab -r 200 -s 1920x1080 -i :0.0 -c copy -f rawvideo -
[19:29:54 CET] <loki_> BtbN: i tried this ffmpeg -f x11grab -r 200 -s 1920x1080 -i :0.0 -c copy -pix_fmt nv12 -f rawvideo - | ./ffmpeg_bin/ffmpeg -f rawvideo -i - -input_format nv12 -r 60 -s 1920x1080 -vcodec libnvenc -f matroska -y /dev/null
[19:30:15 CET] <BtbN> With -c copy the pix_fmt is ignores.
[19:30:17 CET] <BtbN> *d
[19:35:11 CET] <loki_> BtbN: ok now i have just ffmpeg -f x11grab -r 200 -s 1920x1080 -i :0.0 -c copy -f rawvideo - | ./ffmpeg_bin/ffmpeg -pix_fmt bgr0 -f rawvideo -i - -r 60 -s 1920x1080 -vcodec libnvenc -f matroska -y /dev/null
[19:35:44 CET] <loki_> have such output http://pastebin.com/VpHMKyVX
[19:35:50 CET] <BtbN> The codec is called nvenc
[19:36:04 CET] <loki_> it is from not official rep
[19:36:13 CET] <loki_> and it called libnvenc there
[19:36:18 CET] <loki_> i will change it.
[19:36:20 CET] <BtbN> according to your paste, it's an ubuntu build, version 2.4
[19:36:24 CET] <BtbN> this doesn't have either
[19:36:46 CET] <BtbN> But i have no idea what you are trying to achive with that.
[19:36:50 CET] <BtbN> It won't get any faster.
[19:37:54 CET] <loki_> BtbN: http://pastebin.com/sVDmfZEt
[19:38:12 CET] <loki_> BtbN: just working with pipes
[19:38:33 CET] <BtbN> -s and -r propably habe to be in front of -i
[19:40:56 CET] <loki_> ya it is working now
[19:41:03 CET] <loki_> even without pix_fmt
[19:41:27 CET] <BtbN> Well, now the second instance has to do the same conversion
[19:54:29 CET] <loki_> BtbN: i thought i can do such pipe redirect with any loose of perfomance
[19:54:35 CET] <loki_> but things getting worse.
[19:54:42 CET] <loki_> not even close lol
[19:54:45 CET] <BtbN> It's a bunch of extra copies
[19:54:50 CET] <BtbN> of course it gets slower
[19:55:09 CET] <BtbN> it's even a bunch of extra copies for _huge_ amounts of data
[21:03:47 CET] <loki_> BtbN: what is internal ffmpeg pixel format?
[21:04:03 CET] <BtbN> Such a thing doesn't exist.
[21:04:15 CET] <BtbN> It will use whatever it gets
[21:04:26 CET] <BtbN> If the destination doesn't accept it, it will be converted.
[21:10:05 CET] <loki_> BtbN: i don't think that there is simple way to convert from any format to any, specially with cpu optimizations.
[21:10:13 CET] <loki_> so there must be way of doing this.
[21:10:23 CET] <loki_> something standing in between.
[21:11:06 CET] <BtbN> No idea how swscale works internaly, but i'd assume it allways tries to build the most optimal conversion path
[21:11:29 CET] <BtbN> ffmpeg just throws some format at it, and requests what it needs.
[21:12:17 CET] <loki_> is there way to debug this?
[21:12:36 CET] <BtbN> Debug what? There is no way to debug your CPU faster.
[21:14:49 CET] <BtbN> If you want to know what conversion it does you propably just need to raise the log level.
[21:15:12 CET] <loki_> yes.
[21:24:37 CET] <loki_> BtbN: ok, now i understand why not official ffmpeg is little fuster. It uses yuv420p not nv12
[21:25:04 CET] <loki_> i found -loglevel option and that there is ffmpeg_g
[21:25:17 CET] <BtbN> ffmpeg_g is just not stripped from the debug symbols
[21:25:21 CET] <BtbN> nvenc does only accept nv12 though
[21:26:34 CET] <loki_> then how this work then?
[21:26:41 CET] <BtbN> No idea?
[21:27:07 CET] <loki_> BtbN: https://dl.dropboxusercontent.com/u/25725476/screenshot-2015.03.26-23%3A26%3A50.png
[21:27:49 CET] <BtbN> I never looked at that code to avoid license poisoning.
[21:28:11 CET] <loki_> lol
[21:28:50 CET] <BtbN> Not funny at all, they didn't propperly license their code, so just looking at it and implementing stuff in a similar way can be an issue.
[21:29:03 CET] <BtbN> No idea if they ever fixed it.
[21:50:39 CET] <ChocolateArmpits> hello, is anyone familiar with microsoft's smooth streaming ?
[22:46:00 CET] <DragonsLordKERNE> Good night, everybody :)
[23:22:17 CET] <shanlar-> is it possible to set audio as option for a ffserver stream?
[23:22:55 CET] <shanlar-> sometimes the stream has audio and sometimes it doesnt. when it doesnt have audio ffmpeg errors out with "Missing audio stream which is required by this ffm"
[00:00:00 CET] --- Fri Mar 27 2015


More information about the Ffmpeg-devel-irc mailing list