[Ffmpeg-devel-irc] ffmpeg.log.20180907

burek burek021 at gmail.com
Sat Sep 8 03:05:01 EEST 2018


[00:00:42 CEST] <jdel> yeah, V4L2_CAP_VIDEO_CAPTURE is the only thing it checks for
[00:01:26 CEST] <jdel> i suppose just changing that check to look for V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE won't work will it
[00:37:36 CEST] <juny> how can i combine without ID3 metadata
[00:45:48 CEST] <DHE> <barhom> DHE: My 5000$ juniper ex4550 switch has like 4mb of buffer, really sucks   # the QFX 5100 series only has 12 megabytes across all ports, but they have dedicated QoS queues for multicast/broadcast which helps.
[00:46:09 CEST] <DHE> also the 4550 series has given us a lot of problems not related to packet buffers...
[00:49:22 CEST] <barhom> DHE: Interested to know more about what other issues you are referring to. Im a low-level user of these switches, only problems I had was microbursts
[00:53:42 CEST] <DHE> barhom: first we had issues with the SFP+ variant where some modules (notably 1gig RJ-45 modules) would take multiple seconds to do a simple "show interface <...>" which was also reflected in SNMP stat collections. our graphing system basically died to it.
[00:54:12 CEST] <DHE> then we upgraded to some 15.x release and the thing, like, swapped itself to death or something to that effect
[00:54:39 CEST] <DHE> actually no, it wasn't swapping. processes were crashing
[00:54:53 CEST] <DHE> and a few other firmware issues over the years
[00:55:03 CEST] <DHE> maybe they're resolved now but it's left a very bad taste in my mouth
[00:57:39 CEST] <juny> does anyone know how to combine audios without metadata?
[01:17:17 CEST] <juny> I tried with disposition but it doesn't really work
[01:17:31 CEST] <juny> i set disposition 0 but there is still metadata
[01:30:06 CEST] <kevev_> Howdy1
[01:30:08 CEST] <kevev_> HOwdy!
[01:30:10 CEST] <kevev_> blah
[01:30:11 CEST] <kevev_> :p
[01:31:16 CEST] <kevev_> I need to push a single bitrate rtmp feed to a server running ffmpeg and have ffmpeg transcode into multi-bitrate+origin and then have ffmpeg push those streams to a CDN.
[01:31:22 CEST] <kevev_> Anyone know how to do that?
[02:37:03 CEST] <Naan> can someone tell me what's wrong here? https://pastebin.com/8vjfpHsw
[02:38:35 CEST] <poutine> Naan, It looks like a pastebin posting to me without any accompanying errors or description of problems
[02:38:45 CEST] <poutine> -an flag maybe?
[02:39:17 CEST] <Naan> sorry I get "Trailing options were found on the commandline."  and "an output file must be specified"
[02:39:33 CEST] <poutine> oh you don't have anything for the profile
[02:40:11 CEST] <Naan> ya i found it
[02:40:13 CEST] <Naan> >_<
[02:40:24 CEST] <Naan> my bad
[02:41:43 CEST] <Naan> damn what the crap this makes them much bigger
[02:41:50 CEST] <Naan> oh because it has a short interval
[02:47:34 CEST] <poutine> -g is in frames not seconds
[02:48:17 CEST] <Naan> thank you
[02:48:29 CEST] <Naan> i'll use 16 i think
[13:04:50 CEST] <th3_v0ice> Is there a way to detect a live stream using FFmpeg API?
[13:17:52 CEST] <th3_v0ice> Or is there a difference between the regular file and live stream in the API?
[13:36:14 CEST] <michal_f> Hi all
[13:36:14 CEST] <michal_f> I am having a batch job that runs ffmpeg every night. Depending on the computer this runs on (one of many we have) I get mixed results with fonts.
[13:36:14 CEST] <michal_f> Take a look here, please http://www.image-share.com/upload/3831/211.png
[13:36:14 CEST] <michal_f> Looks like some weird font was selected
[13:36:14 CEST] <michal_f> What could I do to fix this? Is there anything like default font for centOS 7 ?
[13:37:49 CEST] <michal_f> this is part of args to ffmpeg: drawtext=text='S12U04 MasterCam Scena__RP v2':fontsize=20:box=1:boxcolor=black:fontcolor=Aqua:alpha=.75
[13:48:13 CEST] <Nacht> Looks like www.johnvansickle.com got himself a bad cert
[13:51:34 CEST] <BtbN> works fine for me, and if it was, he's not here anyway.
[13:51:42 CEST] <michal_f> ok - I just pointed to font on a shared network folder
[13:52:41 CEST] <Nacht> Hmmm. Looks like Firefox has distrusted his cert (or CA). WOrks in Chrome though
[13:53:40 CEST] <BtbN> What's firefox complaining about?
[13:56:41 CEST] <furq> SEC_ERROR_REVOKED_CERTIFICATE
[14:01:39 CEST] <BtbN> Well, Chrome probably hasn't picked that up yet then
[15:03:44 CEST] <th3_v0ice> Can I safely use protocol_whitelist in AVFormatContext to tell whether or not a stream is live  or not?
[15:04:49 CEST] <DHE> maybe, maybe not. I can provide an HTTP source that is live and an HTTP source that is pre-rendered
[15:08:06 CEST] <th3_v0ice> DHE: Thanks for yesterdays response, AVFormatInput was the answer. Onto this question... What if I only wanted to make a difference for example between the live camera and a file, /dev/video0 and /home/user/file
[15:08:56 CEST] <th3_v0ice> Is there any variable that would tell me so or give a hint?
[15:10:22 CEST] <DHE> I think your best bet is to see if you get a sane value in AVFormatContext->duration once it's opened and working
[15:10:48 CEST] <DHE> it's still no guarantee, but if ffmpeg can provide a length you can probably assume it'll shutdown at the end of it
[15:11:44 CEST] <th3_v0ice> Hmm, I will look into it. Thanks!
[18:59:54 CEST] <KomiaPoika> hi
[19:00:42 CEST] <KomiaPoika> how do i use ffmpeg to transform a 6 channels surround, 48khz 32bit wav file to a stereo 44khz 16bit wav file for burning cd?
[19:02:52 CEST] <durandal_1707> ffmpeg -i input.wav -ac 2 output.wav
[19:05:06 CEST] <KomiaPoika> durandal_1707: and to fix khz rate without changing playback speed?
[19:08:54 CEST] <durandal_1707> ah, ffmpeg -i input.wav -ac 2 -af aresample=44100 output.wav
[19:09:39 CEST] <KomiaPoika> thanks
[20:13:41 CEST] <nicolas17> I might be asking something unrealistic but...
[20:15:27 CEST] <nicolas17> I have a ton of JPEG files from a timelapse, they take a lot of space, how can I compress them without losing *any* quality? I tried x264 with -crf 0 and the result was bigger, I expected it to be smaller even when lossless because it would take advantage of temporal redundancy
[20:19:04 CEST] <DHE> you'd think so, but x264 is going to do stupid things like preserve the jpeg artifacts.
[20:21:40 CEST] <kepstin> the best you can probably do is use an appropriate crf setting to get minimal/unnoticable quality loss, and the result should be smaller then.
[20:22:15 CEST] <kepstin> presumably with lossless any temporal redundancy between frames is completely overwhelmed by noise or distortion.
[20:22:59 CEST] <DHE> or maybe x264 special encoding option nr=200 or so? (raise higher for large images, like 720p and beyond)
[20:26:15 CEST] <nicolas17> my originals are 2592x1944 JPEGs, I wanted to make it take less space but still keep good quality to do further transcodings from that, that's why my first attempt was using lossless, but looks like that's not a good idea :P
[20:27:16 CEST] <DHE> while nr=1000 isn't lossless, it might be acceptable. that or try near-lossless with a crf of like 2-5
[20:27:58 CEST] <nicolas17> btw, at that resolution preset=veryslow gives me 1fps encoding lol
[20:28:16 CEST] <DHE> sounds about right
[20:28:22 CEST] <kepstin> well, it says "very slow" right on the box :)
[20:30:21 CEST] <nicolas17> 2.4GB of RAM too
[20:31:13 CEST] <DHE> that's something that's always confused me. look-ahead for VBR mode makes sense, but in crf does it really make that much sense to buffer multiple seconds of video in memory?
[20:33:06 CEST] <nicolas17> maybe it's buffering lots of potential reference images to do prediction from?
[20:33:57 CEST] <JEEB> yes
[20:34:22 CEST] <nicolas17> https://paste.kde.org/pdv0s3vls/n0yias/raw
[20:34:22 CEST] <JEEB> it's not only to enable rate control to take peeks (rc-lookahead), but to also check various coding related things
[20:34:47 CEST] <DHE> well reference frames are limited to 15 (?) but I've seen lookahead of 60
[20:35:26 CEST] <JEEB> I can't be arsed to show the technical details to be honest :P
[20:35:40 CEST] <nicolas17> is it a fixed set of 15? or could the encoder be picking which 15 of the 60 frames is best to use?
[20:35:42 CEST] <JEEB> but tl;dr it does various coding optimization as well as RC stuff
[20:35:45 CEST] <DHE> I'll take your word for it that there's a practical reason. I just can't think of one myself.
[20:35:55 CEST] <JEEB> nicolas17: yea it has been adaptive for a very long time
[20:36:11 CEST] <DHE> nicolas17: it's a spec limit. h264 allows temporal references up to 15 frames backwards. older codecs like mpeg4 (divx) only used the previous frame (1 reference)
[20:36:37 CEST] <JEEB> yes but you still can put different coded picture numbers in there AFAIK
[20:36:48 CEST] <JEEB> so it's not like it has to be "the last 15 frames"
[20:36:50 CEST] <nicolas17> max 15 different frames potentially far away, vs max 15 frames away
[20:37:02 CEST] <DHE> one frame may reference multiple previous frames
[20:37:15 CEST] <nicolas17> if h264 allows the former then it makes sense the encoder could buffer lots
[20:37:29 CEST] <DHE> oh, I thought it was only the last decoded 15 frames available for reference
[20:37:40 CEST] <nicolas17> pastebin above shows the filesizes I'm getting
[20:37:54 CEST] <JEEB> DHE: nah
[20:37:58 CEST] <JEEB> pretty sure it's just 15 indices
[20:37:59 CEST] <nicolas17> mjpeg.mkv is "-vcodec copy -f matroska"
[20:38:06 CEST] <JEEB> so that 15 frames have to be kept in memory
[20:38:11 CEST] <JEEB> then they get updated every now and then
[20:38:36 CEST] <kepstin> Hmm. Does the encoder have to signal that a particular frame will be used as a ref later, to tell the decoder which frames to keep?
[20:38:45 CEST] <nicolas17> I wonder how well x265 would work! *my CPU starts crying*
[20:38:53 CEST] <JEEB> kepstin: I think that's how it works
[20:39:06 CEST] <DHE> nicolas17: you might get 30-40% off the file size, but yes there will be much crying
[20:39:13 CEST] <JEEB> for lossless?
[20:39:28 CEST] <JEEB> nicolas17: remember that if you don't need seeking then just use keyint infinite
[20:39:28 CEST] <kepstin> nicolas17: honestly, hard drives are cheap, the jpeg's aren't really that large, just keep them around as the master copy :/
[20:39:50 CEST] <JEEB> paired with scenecut=0
[20:39:51 CEST] <DHE> and at 300 megabytes... that's nothing. USB sticks bigger than that cost $3
[20:40:00 CEST] <JEEB> so that x264 never makes another IDR
[20:40:10 CEST] <nicolas17> DHE: this is 180 frames for testing so I don't have to wait hours for a filesize comparison
[20:40:16 CEST] <DHE> ah....
[20:40:43 CEST] Action: DHE checks one of his own time lapses...
[20:40:44 CEST] <JEEB> but yea, if your input is already lossy and the pictures can't be used too well for prediction?
[20:41:01 CEST] <JEEB> if they can, then just stick keyint infinite and scenecut=0 unless you need seeking
[20:41:18 CEST] <DHE> yeah that's why we're suggesting he take some minimal lossy options like crf=2 just to get the file size down
[20:42:07 CEST] <nicolas17> https://cdn.discordapp.com/attachments/286612533757083648/487688979882311720/sample.mp4 (same 180 frames, -vf scale=1024:768 -preset slow -crf 23 to show what the content is like)
[20:42:14 CEST] <DHE> x264 option nr=1000 or 2000 might be worth trying as well. but do examine it...
[20:42:52 CEST] <nicolas17> kepstin: I think at this point it stopped being a practical question and turned into experimenting with x264 lol
[20:43:17 CEST] <nicolas17> ie. I might keep playing with options out of curiosity even if I'm going to keep the JPEGs
[20:45:42 CEST] Action: DHE has 96 GB of PNGs for a time lapse of over 110,000 frames
[20:46:44 CEST] <nicolas17> DHE: this is straight off a GoPro, about 1.8MB per photo
[20:47:27 CEST] <DHE> mine was taken out of a live video feed, ffmpeg grabs 1 frame on a cron job every 5 minutes
[20:47:42 CEST] <DHE> it's a building being constructed. the video spans multiple months.
[20:48:00 CEST] <nicolas17> ohh neat
[21:42:26 CEST] <nicolas17> when encoding an image sequence, is there any way I can hardsub the filename?
[21:42:57 CEST] <nicolas17> the 'drawtext' videofilter has a way to use 'frame metadata' in the text, but is the frame filename kept in any metadata key?
[21:57:35 CEST] <th3_v0ice> How do I read a pipe with API? Setting the format and providing the pipe name only doesnt work :)
[23:18:21 CEST] <th3_v0ice> Do I need to follow some other protocol or should I just use avformat_open_input() and then avformat_find_stream_info()?
[23:33:56 CEST] <DHE> th3_v0ice: that's usually enough unless you have some specific needs
[23:44:24 CEST] <pauletin> is the native aac encoder considered to be better than libfdk_aac now?
[23:46:24 CEST] <BtbN> no
[23:46:31 CEST] <BtbN> But not notably worse anymore
[23:48:17 CEST] <pauletin> at 128 and 192 kbps, which encoder is better? Also, VBR only works for libfdk_aac as I know, but is it recommended?
[23:48:36 CEST] <pauletin> the trac guide seems to be 2 years out of date
[23:51:43 CEST] <ChocolateArmpits> libfdk_aac has a max cutoff at 20kHz, if you want to preserve everything then it's probably better to use the aac encoder
[23:52:09 CEST] <JEEB> pauletin: 128 or 192kbps is a rather high bit rate anyways (for stereo)
[23:56:41 CEST] <th3_v0ice> DHE: Thanks, stupid question on my side, its working fine with just the plain path. Another question, is there a way to determine the number of samples in audio frame in mpegts stream? frame_size is 0.
[23:57:15 CEST] <furq> pauletin: recommended for what
[23:57:27 CEST] <th3_v0ice> DHE: codec is AC3
[23:57:27 CEST] <furq> if you're asking if it works well then yeah
[23:57:39 CEST] <furq> -vbr 5 should be transparent
[23:58:07 CEST] <JEEB> th3_v0ice: you only get samples from audio frame after decoding
[00:00:00 CEST] --- Sat Sep  8 2018


More information about the Ffmpeg-devel-irc mailing list