[Ffmpeg-devel-irc] ffmpeg.log.20160314

[burek]

[04:45:35 CET] <jookiyaya> anybody have ffmpeg build that support  aac-fdk and mp3 shine
[04:47:46 CET] <furq> stop asking people to violate the gpl
[04:48:25 CET] <furq> if you want an ffmpeg build with fdk support, build it yourself
[04:49:13 CET] <jookiyaya> i have no idea how to
[04:49:35 CET] <jookiyaya> and mp3 shine doesn't work either
[04:51:14 CET] <furq> well then learn how to or use a different aac encoder
[04:51:16 CET] <jookiyaya> and  what version # is this:   R:\>ffmpeg64 --version
[04:51:16 CET] <jookiyaya> ffmpeg version N-78843
[04:51:39 CET] <jookiyaya> furq i bet you don't even know how to either
[04:57:42 CET] <c_14> jookiyaya: there should be a -g<something> after that. that something is the git commit hash
[04:59:54 CET] <jookiyaya> https://ffmpeg.zeranoe.com/builds/win64/static/ffmpeg-3.0-win64-static.7z
[05:00:02 CET] <jookiyaya> what does it mean by "shared version"
[05:00:07 CET] <jookiyaya> vs "static version"
[05:01:10 CET] <c_14> The shared version has shared libraries, the static version has static libraries
[05:01:39 CET] <jookiyaya> what does it mean by having 'shared libraries" and "static libraries"
[05:03:25 CET] <PrestonL> Anyone know of a module or way for there to be an XML or JSON stat.html page so that another application can query the server status and display that info in another way?
[05:04:11 CET] <c_14> jookiyaya: http://cs-fundamentals.com/tech-interview/c/difference-between-static-and-dynamic-linking.php
[05:04:58 CET] <jookiyaya> does it make a difference in final encoding product
[05:05:09 CET] <jookiyaya> whether you used static or shared
[05:05:20 CET] <c_14> no
[05:05:41 CET] <jookiyaya> from which ffmpeg version did they removed fdk-aac
[05:06:01 CET] <c_14> none
[05:06:38 CET] <jookiyaya> i was told they had it before
[05:06:44 CET] <c_14> It was never in a distributible build (but you can still compile ffmpeg with it yourself)
[05:22:54 CET] <jookiyaya> what is correct file extension for  audio-only aac file
[06:21:43 CET] <jookiyaya> is there 2 different vorbis encoder in ffmpeg?
[10:43:43 CET] <eightfold> hi there! i have a video that is 720x576px. i want to upscale it to 1920x1080 and put black borders on the left and right sides.
[10:50:03 CET] <eightfold> i've used:  ffmpeg -y -i input.mp4 -aspect 5:4 -vf "scale=1920:1080" output.mp4
[10:50:26 CET] <eightfold> but when i check media information in vlc it says "resolution: 1920x1090"
[10:50:37 CET] <eightfold> and "display resolution: 1920x1080"
[10:51:55 CET] <eightfold> can this be fixed? input files also says: "resolution: 720x578" and "display resolution: 720x576"
[11:24:49 CET] <maxinminax> i need compile ffmpeg with ocr filter (use tesseract lib) but when do that, it have errror "ERROR: tesseract not found using pkg-config"
[11:25:20 CET] <maxinminax> anyone can help me?
[11:25:38 CET] <maxinminax> or anyone have pre-build version which have build options --enable-libtesseract
[11:36:41 CET] <flux> is there some modern video format (supported by ffmpeg) that's playable while the file is being written? mpeg4 does not seem to fit the bill :)
[11:37:12 CET] <flux> it also seems ffmpeg cannot write the required header information without releasing memory etc, so a program cannot just write the headers once a minute or so
[11:38:58 CET] <Mavrik> flux, MPEG-TS can be read while written
[11:39:12 CET] <Mavrik> It's used everywhere where live streaming is required for that reason ;)
[11:54:09 CET] <Bluez_> hi
[12:40:32 CET] <flux> mavrik, can I seek it with ffmpeg-based players?
[12:44:01 CET] <flux> actually I think I can, though it might not work if I seek beyond the track length that existed when I start the player
[12:44:11 CET] <flux> but yeah, MPEG-TS is probably the way to go. thanks!
[12:44:40 CET] <flux> (and once the save is finished (I split it in one-hour chunks) I can just convert it to MPEG4)
[12:46:20 CET] <Mavrik> MPEG4 is not a container
[12:46:22 CET] <Mavrik> :)
[12:46:55 CET] <flux> fine, ISO Media File Format then ;)
[12:47:25 CET] <flux> actually is MP4 not a container format..
[12:48:12 CET] <Mavrik> MP4 is a container format but it's a tiny part of the spec :)
[12:48:24 CET] <Mavrik> MPEG-TS is also part of MPEG spec etc. :)
[12:48:29 CET] <Mavrik> That's why it can be confusing :)
[12:49:30 CET] <flux> however, copying H264-encoded data from MPEG-TS to an MPEG4 is a completely sensible thing to do, no?-(
[12:49:42 CET] <Mavrik> Well, depends :)
[12:49:52 CET] <Mavrik> It can be done easily and quickly.
[12:50:11 CET] <BurnerGR> I have troubles getting a decent quality when live encoding 1080p at 60 using libx264, it does not seem to utilize much CPU regardless of the number of --threads, it just drop fps when increasing quality settings
[12:50:28 CET] <Mavrik> BurnerGR, pastebin command + output if you can
[13:20:57 CET] <BurnerGR> Mavrik, this is an example where encoding fps drops, but cpu load is low http://pastebin.com/z2r6rA9a
[13:28:58 CET] <BurnerGR> Mavrik, oh, looking at my own paste, I see that cpu load is not low, ffmpeg is  just "niced"
[13:38:32 CET] <DHE> BurnerGR: libx264 sets some of its own encoding threads as low priority
[13:39:03 CET] <DHE> also you're mixing -b:v with -crf
[13:39:45 CET] <DHE> it does look like you're CPU-bound. maybe try veryfast?
[13:40:12 CET] <BurnerGR> veryfast works, but its a horrible quality
[13:41:55 CET] <BurnerGR> I'll assign some more cores and see if it helps
[13:56:31 CET] <DHE> oh it's a VM
[13:57:54 CET] <DHE> I'm inclined to say that for 60fps 1080p, you might want to raise the bitrate a little bit. depending on the media 4.5 megabits is a bit on the low side, possibly better suited for 30fps material
[14:08:20 CET] <BurnerGR> DHE, media source is mixed, and can have a lot of action (fps gameplay), there is no problem with quality during interviews and such
[14:08:46 CET] <DHE> ah, video games. sharp edges. yeah that'll be hard
[15:14:17 CET] <Carl___> I have a problem with ffserver. I can feed the server in linux and play in windows. But i can't feed it under windows
[15:14:43 CET] <Carl___> the command i'm using is: ffmpeg -i sample.mpg http://192.168.10.117:8090/feed.ffm
[15:51:14 CET] <wizonesolutions> OS X: I'm trying to record system audio on Mac, and I've installed iShowU Audio Capture and set up a multi-output device to get system audio into a device usable by ffmpeg when using the avfoundation format. However, the audio comes in faster than the video, and I'm not sure where to start trying to troubleshoot that.
[15:51:36 CET] <wizonesolutions> Or if something like Soundflower actually *would* be better (googling implies it doesn't work that great)
[15:59:10 CET] <jarainf> "Audio comes in faster"?
[15:59:21 CET] <jarainf> Is it delayed?
[16:05:13 CET] <xeons> Having audio / video sync issues do to timestamps and offsets when I -ss crop out sections of h.264 video: http://stackoverflow.com/questions/35416110/ffmpeg-concat-video-and-audio-out-of-sync/
[16:05:56 CET] <xeons> I've tried apad, -c copy, -shortest, -avoid_negative_ts, make_zero, -fflags, +genpts, etc.. without luck
[16:06:40 CET] <durandal_1707> apad adds audio at end
[16:06:52 CET] <xeons> Yes, I was using it to try to make the audio last as long as the video
[16:07:08 CET] <xeons> + -shortest
[16:07:33 CET] <durandal_1707> you can't use it with copy
[16:07:53 CET] <xeons> Good to know. I've tried it with an encode and it failed too.
[16:08:54 CET] <xeons> I can do better than that. Here is the source video and commands: http://davidpennington.me/share/audio_sync_test_video.zip
[16:13:36 CET] <Carl___> I'm getting an error
[16:13:38 CET] <Carl___> HTTP error 503 Server too busy
[16:13:47 CET] <Carl___> http://192.168.10.117:8090/feed1.ffm: Server returned 5XX Server Error reply
[16:13:51 CET] <Carl___> How can i fix that?
[16:17:21 CET] <Carl___> anyone?
[16:29:01 CET] <lakis> Hello! Anyone have any examples of batch processing with the CLI? I have big folder with lots of subfolders of mostly mp3's and would like to batch convert them to small m4a/aac files
[16:31:35 CET] <andrey_utkin> lakis, just traverse your folders in a loop, you can do that in any programming language you like
[16:31:58 CET] <andrey_utkin> ffmpeg doesn't process many files at once in your case
[16:34:53 CET] <lakis> yeah, i was looking to save some time thinking that someone must have done something similar before. I need a linux script that  picks all mp3's and wma's in all subfolders and re-encodes them into smaller sized m4a's in place. Can't find any good examples online, have any links?
[16:35:39 CET] <lakis> I guess something similar to this, for audio: for f in *.avi; do ffmpeg -i "$f" -c:v libx264 -crf 23 -preset medium \   -c:a libfdk_aac -vbr 4 -movflags +faststart -vf scale=-2:720,format=yuv420p \   encoded/"${f%.avi}.mp4"; done
[16:36:00 CET] Last message repeated 1 time(s).
[16:38:17 CET] <furq> find . \( -name "*.mp3" -o -name "*.wma" \) -exec ffmpeg -i "{}" ... "{}.m4a" \;
[16:39:07 CET] <Carl___> Why does this wait_feed problem occurs?
[16:39:12 CET] <Carl___> I thought it was repaired
[16:39:36 CET] <furq> i wouldn't hold your breath for help with ffserver, it's practically abandoned
[17:55:20 CET] <n1cky_> Can anyone shed some light on how I can use ffmpeg to buffer an http stream, such that I can rewind it ~60 seconds and make recordings?
[17:56:03 CET] <n1cky_> The end goal is to be able to make short clips after something interesting happens in the stream.
[17:56:39 CET] <teratorn> is there an easy way to say, cut the last 5 seconds off a video without having to do manual calculations? Ie. with a -5s parameter somewhere or something like that?
[17:58:26 CET] <n1cky_> teratorn: man ffmpeg-utils "Time duration"
[17:58:49 CET] <n1cky_> haven't tried it so I don't know, but there is a "optional - indicates negative duration."
[17:59:37 CET] <teratorn> also whats a good ffmpeg front-end that runs on os x? :)
[17:59:43 CET] <teratorn> n1cky_: thanks ill check it
[18:11:37 CET] <flux> mavrik, ts works just out-of-the-box with my app, but apparently mpv gets easily confused by time stamps that don't start at 0. ffplay works fine, though.
[18:12:13 CET] <flux> (even mpv works ok if said time stamps are inside an mp4)
[19:02:37 CET] <Zucca> Do i need to specify -threads _after_ input files to make it function? I'm trying to encode vp9 video, but I'm only getting "one core hot".
[19:03:15 CET] <BtbN> you don't need to specify it at all if you want it to use all cores.
[19:03:17 CET] <DHE> yes, -threads is processed on a per-stream basis
[19:03:40 CET] <Zucca> DHE: Thanks.
[19:04:34 CET] <Zucca> So I guess if I specify it before inputs, it gets applied to decoding process.
[19:05:30 CET] <DHE> but BtbN's point stands. unless you have specific requirements, the defaults are to autodetect number of threads to use
[19:06:22 CET] <BtbN> libvpx isn't exactly good in using the CPU though
[19:06:31 CET] <Zucca> Doesn't make sense. O.o If I specify threads before inputs I get one thread.
[19:06:54 CET] <Zucca> So better not to specify it at all.
[19:06:56 CET] <BtbN> no, that doesn't make sense. You get number-of-cores threads if you don't speficy it for the encoders.
[19:19:25 CET] <Zucca> Dang...
[19:19:38 CET] <Zucca> Anyway I do it, I get only one thread.
[19:19:47 CET] <Zucca> At least one active.
[19:20:48 CET] <kepstin> Zucca: it depends a bit on other encoding options, frame size, etc. whether having more threads is actually usable or will help
[19:23:20 CET] <Zucca> I looked for the options from here: http://wiki.webmproject.org/ffmpeg/vp9-encoding-guide
[19:24:02 CET] <Zucca> Multi-threaded encoding may be used if -threads > 1 and -tile-columns > 0.
[19:24:26 CET] <Zucca> Last Modified: 2015-09-08
[19:25:00 CET] <Zucca> So I don't know if the information there is already obsolote/faulty.
[19:27:29 CET] <kepstin> what size of video are you using? iirc, multithreaded vp9 has the biggest gains when you're doing HD or 4K video sizes. Playing around with the '-speed' setting might make a difference.
[19:27:47 CET] <kepstin> could try explicitly setting -frame-parallel 1 as well.
[19:29:09 CET] <Zucca> kepstin: -pass 1 -frame-parallel 1 -tile-columns 6 -speed 4 ... 1080p
[19:29:28 CET] <Zucca> I wonder it -pass 1 makes the difference.
[19:34:02 CET] <Zucca> Nope.
[19:37:41 CET] <atomnuker> don't use frame-parallel, it reduces compression and using ffmpeg to decode actually reduces performance since it doesn't fully support that mode
[19:38:04 CET] <atomnuker> IIRC it only works fine with libvpx
[19:38:26 CET] <Zucca> I'm decoding from lossless format and encoding to vp9+opus, webm.
[19:38:45 CET] <kepstin> hmm, fun. with x264, frame-parallel has long been the preferred mode :/
[19:39:43 CET] <kepstin> Zucca: be interesting to compare the performance of doing only the decode vs. doing the decode+encode. Some lossless codecs can be pretty heavy to decode iirc.
[19:40:38 CET] <Zucca> kepstin: I'm using zlib video as input here.
[19:41:22 CET] <Zucca> It's pretty tasking to decode, but I can playit real-time just fine.
[19:42:50 CET] <Zucca> htop shows about 133% CPU usage in real time decoding. 600% being maximum (hexa-core CPU).
[19:43:56 CET] <kepstin> hmm, so you're already saturating a single thread doing only the decoding. ffmpeg's core is single-threaded, so you might be hitting some bottlenecks due to the decoding? not sure.
[19:45:00 CET] <Zucca> kepstin: input fps 35 and I'm gertting max of 1 fps encoding.
[19:45:37 CET] <Zucca> With 1 fps decoding surely isn't the bottleneck. :\
[19:46:09 CET] <kepstin> but yeah, if doing frame threading in libvpx isn't recommended, the only other option would be slice threading, which of course *also* reduces the compression.
[19:46:31 CET] <atomnuker> the frame-parallel variable for libvpx actually refers to decoding each tile in parallel, rather than each frame in parallel
[19:46:45 CET] <kepstin> hmm. not sure how vp9 works actually
[19:46:50 CET] <kepstin> i've only ever done vp8 stuff
[19:46:59 CET] <atomnuker> ffmpeg supports only doing frames in parallel and will not do per-tile decoding in separate threads
[19:47:42 CET] <Zucca> I removed it. Still getting one active thread. With -speed 4 I get maximum of 2.2 fps encoding with audio dropped out.
[19:48:00 CET] <kepstin> Zucca: what's the video size?
[19:48:04 CET] <kepstin> oh, 1080p
[19:48:26 CET] <kepstin> if vp9 is doing some parallelization (per tile, i guess?) that should be big enough to show it
[19:49:24 CET] <kepstin> but yeah, both vp9 and x265 are sort of 'next generation' codecs, and as such are both very slow
[19:50:05 CET] <Zucca> Now there's something wrong... with vp8 (libvpx) and same settings I get 3.5 fps and one active thread.
[19:50:10 CET] <kepstin> (or alternatively, if tuned to be the same speed as current generation codecs, don't really provide much in the way of quality improvements)
[19:51:11 CET] <kepstin> Zucca: yes, vp9 is slower than vp8 at encoding :)
[19:52:01 CET] <Zucca> kepstin: Yeah. I know. But I cannot get multithreaded encoding with vp8 either. And its speed is also _slow_.
[19:53:22 CET] <kepstin> i seem to recall that libvpx-vp8 had worse multithreading ability than vp9 :/
[19:54:38 CET] <Zucca> I guess it's still libx264 that's most balanced in speed and quality. :(
[19:55:13 CET] <Zucca> I just wished I could get bit better performance aout of thise free vp codecs. :|
[19:55:30 CET] <kepstin> are you using "-quality good"? I forget what the default is, but you generally don't want to use "best" (it's akin to x264's placebo preset in that it makes it slower without any real gain in quality)
[19:57:02 CET] <Zucca> My full cli line was/is: ffmpeg -y -i in.mkv -acodec opus -ab 48k -vcodec libvpx-vp9 -b:v 330k -tile-columns 6 -auto-alt-ref 1 -lag-in-frames 25 -speed 4 out.webm
[19:57:29 CET] <Zucca> I've tried with -speed 1 also.-
[19:57:45 CET] <Zucca> But that just slows it down (obviously).
[19:57:48 CET] <kepstin> 330k for a 1080p video? that's probably not gonna give very good results...
[19:58:04 CET] <Zucca> kepstin: Yeah. This it just for testing.
[19:58:52 CET] <kepstin> how doe the performance compare when using a single-pass encode with -crf option set to something resonable?
[19:59:01 CET] <Zucca> Giving it 3M doesn't affect the speed. :(
[19:59:06 CET] Action: kepstin wouldn't expect much difference tho
[19:59:31 CET] <Zucca> Let's try.
[20:00:58 CET] <kepstin> (in general, i'd expect using -pass 1 shouldn't make the encode slower; in some codecs it can make it faster by disabling parts of the encoder that aren't needed for the first-pass bitrate analysis)
[20:01:00 CET] <Zucca> Under 2 fps and single threaded.
[20:01:23 CET] <Zucca> What it happening here..? :D
[20:01:57 CET] <Zucca> I could try to compile libvpx instead of using prebuilt...
[20:02:57 CET] <kepstin> what kind of cpu is it anyways? There might be some contention issues on AMD processors - video encoding is heavily using floating point, and every 2 "cores" on the amd chips share floating-point resources.
[20:03:34 CET] <kepstin> although I'd still expect it to show up in top as using multiple cores
[20:04:30 CET] <kepstin> but yeah, my general experience with the vpx codecs has also been "kinda slow and don't parallelize very well" :/
[20:06:17 CET] <kepstin> (even x264's frame-based threaded encoding mode isn't ideal in some cases; it reduces coding efficiency since it limits the area usable for reference, and adds latency such that it's not useful for "realtime" stuff)
[20:06:32 CET] <Zucca> kepstin: I have  AMD Phenom II X6 1055T 2.8GHz. The architecture in this CPU isn't like on those FX CPUs. Those six core don't share caches.
[20:07:46 CET] <Zucca> I checked this from cpuworld. Each core is "independent".
[20:08:07 CET] <kepstin> Zucca: yeah, i see why you want multithreading to work, that's a pretty slow/old cpu by modern standards :/
[20:08:14 CET] <jkqxz> Floating point what?  Except for some legacy things, video encoding is entirely fixed point.
[20:09:00 CET] <Zucca> kepstin: But you might be right about this... Since few of my friends managed to get multi-threaded encoding - they all have intel CPUs.
[20:09:46 CET] <kepstin> jkqxz: ok, not floating point, but with assembly heavy use of the vector units which usually share resources with floating point hardware
[20:10:37 CET] <Zucca> I'm gonna upgrade to ocata-core 4GHz FX CPU... But it just might make things worse because of the shared caches. Although all the specs and benchmarks indicate 20-60% boost in performance...
[20:13:50 CET] <Zucca> One _really_ dirty thing to do would be to split the input and then run _seperate_ ffmpeg processes to encode raw streamable vp9 data. Fianlly stream copy all (concatenate) all the pieces and encode audio into a container. :P
[20:14:37 CET] <Zucca> That would need some scripting. And woudn't be very good in compression...
[20:14:54 CET] <Zucca> But hey! Doable! (Won't do it anyway).
[20:16:31 CET] <Zucca> I did once a script that runs multiple optipng processes to speed up compressing multiple (or big in size) pngs.
[20:16:41 CET] <Zucca> It was fun.
[20:19:56 CET] <kepstin> Zucca: if you do it at intervals which are significantly longer than the keyframe interval, the compression loss would be somewhat minimal (assuming you're using "crf" mode for a quality based encode)
[20:20:11 CET] <kepstin> obviously, two-pass bitrate encodes are out if you do that
[20:22:11 CET] <Zucca> Yeah. Or well. You can use them, but the quality between chucks/parts might change very much.
[20:23:36 CET] <Zucca> But yeah. Doing it using crf and seperate processes on a video file that's long enough... why not. I'm not just going to go there right now.
[20:56:25 CET] <daslicht> hi,
[20:56:49 CET] <daslicht> which data of a mp3 file do I need to be able to determine its bytes / second so that I can iterate over it in 1 second steps ?
[20:58:35 CET] -daslicht:#ffmpeg- why does ffpmeg return 2 differnet bitrates for one file ?
[20:58:35 CET] -daslicht:#ffmpeg- https://gist.github.com/daslicht/3d7e85fa2345655152b0#file-gistfile1-txt-L27
[20:58:35 CET] -daslicht:#ffmpeg- https://gist.github.com/daslicht/3d7e85fa2345655152b0#file-gistfile1-txt-L52
[21:17:54 CET] <kepstin> daslicht: one's from the container, the other's from the codec. In this case, it looks like the format one is just "filesize (bytes) / duration * 8" so it could be office, since it'll include the metadata headers, etc.
[21:18:10 CET] <kepstin> so it could be off by a bit*
[21:50:22 CET] <needmorespeed> I'm not finding any useful settings for H264 decoding speed. I've tried thread_count, doesn't seem to make a difference. Any suggestions?
[21:51:20 CET] <TD-Linux> there aren't really any settings for it
[21:51:59 CET] <kepstin> needmorespeed: it already decodes as fast as possible by default, since why would anyone want to make it slower?
[21:52:02 CET] <Carll_> I would like to feed ffserver with some encoded packets.I'm using " av_interleaved_write_frame( format_out_ctx, outPacket ); " function to write a packet to an output media file. I can connect with the server with avformat_write_header( format_out_ctx, NULL ) function, and it is working fine (i think). If I pass that function, the server shows me " [POST] "/feed1.ffm HTTP/1.1" 200 4096 ""
[21:52:06 CET] <Carll_> But my problem is, the av_interleaved_write_frame() function doesn't feed the server.
[21:52:08 CET] <Carll_> Can anyone tell me what am i doing wrong? If somebody is willing to help me, I could send ffserver.conf and my C files.
[21:53:32 CET] <kepstin> needmorespeed: if you can't cpu decode the video fast enough, you might want to look at a hardware decoder (vaapi, vdpau, qsv, whatever)
[22:11:13 CET] <Carll_> someone?
[22:21:29 CET] <Carll_> Can someone help me please? The following command is working "ffplay -rtsp_flags listen -i rtsp://127.0.0.1:8090/live.sdp". It streams my desktop.
[22:21:37 CET] <Carll_> But i would like to feed the server
[22:21:50 CET] <Carll_> but it doesn't work
[23:49:50 CET] <Dan0maN> hello all.  i have an IP camera that writes .jpg stills into my NAS.  wanting to create an mpeg out of them.  i found an example, but not working:  http://www.itforeveryone.co.uk/image-to-video.html
[23:50:31 CET] <Dan0maN> creating paste.  one sec
[23:52:17 CET] <Dan0maN> https://gist.github.com/anonymous/332c77ee54ffd25114a7
[23:52:59 CET] <Dan0maN> any help is appreciated.  completely noob at ffmpeg
[23:54:11 CET] <furq> Dan0maN: https://www.ffmpeg.org/ffmpeg-formats.html#Examples-1
[23:55:13 CET] <llogan> Dan0maN: you overwrote your inputs
[23:57:05 CET] <Dan0maN> k.  yeah, at least it prompted me, which why i copied them into a new directory first
[23:57:10 CET] <Dan0maN> ty.  i'll read up
[23:58:59 CET] <Dan0maN> okay.  i thought that %05d was an http encoding thing, but i guess those actually mean something to the cli?
[23:59:42 CET] <llogan> 00001, 00002, 00003, etc
[23:59:52 CET] <Dan0maN> yeah.  trying to search for which one to use
[00:00:00 CET] --- Tue Mar 15 2016