[Ffmpeg-devel-irc] ffmpeg.log.20180719

[burek]

[04:04:33 CEST] <jnollette> anyone notice some processors perform way less with ffmpeg
[04:04:59 CEST] <jnollette> three 4790s. one gets 1/2 the peformance
[04:05:48 CEST] <DHE> with the exact same binary version, input file and commandline?
[04:06:06 CEST] <jnollette> yes
[04:06:33 CEST] <jnollette> maybe 3/4 of the average
[04:06:42 CEST] <DHE> could be lots of things. systems that are busy or other bottle necks, BIOS and CPU clock settings, memory channels
[04:07:00 CEST] <DHE> 2 sticks of RAM (properly inserted of course) can give around 30% performance boost over 1 stick
[04:07:22 CEST] <jnollette> only difference is motherboard and ram brands
[04:24:13 CEST] <zx_> en?
[04:24:26 CEST] <zx_> this can not save the chat history ?
[04:26:01 CEST] <nicolas17> zx_: this particular channel is publicly logged: https://lists.ffmpeg.org/pipermail/ffmpeg-devel-irc/2018-July/005150.html
[05:58:35 CEST] <kepstin> intel cpus throttle themselves based on power/heat, so a poorly installed hsf can do that sort of thing
[05:59:28 CEST] Action: kepstin would suggest using the 'turbostat' tool while ffmpeg is running to check clock speeds and temperatures.
[09:54:21 CEST] <th3_v0ice> What is the best way to generate audio PTS and DTS? Is this correct way to do it: pts = dts = ((1/(fps.num/fps.den)) * timebase.den * count) / frame->nb_samples?
[09:54:58 CEST] <Mavrik> Why would you base audio pts/dts on frames?
[09:55:03 CEST] <Mavrik> Audio doesn't really have fps
[09:55:16 CEST] <Mavrik> If you have a track that's not timestamped, use sample count and samplerate to timestamp them
[09:58:03 CEST] <th3_v0ice> I need to sync it with video
[09:58:36 CEST] <Mavrik> So you're going to have gaps where audio cuts out and breaks players in between? :)
[10:11:40 CEST] <th3_v0ice> I am not following you. This frame is an audio frame, and it has 1024 samples in it. Wouldnt that mean that in 24fps sequence, this audio frame lasts 0.04 seconds?
[10:13:32 CEST] <JEEB> samples/sample_rate amount of time
[10:17:32 CEST] <th3_v0ice> Ok, thanks guys
[10:25:38 CEST] <automatical> Hey folks, I'm consuming a TCP stream that has a fairly unreliable frame rate, then pushing this to a live streaming service, is there a way to have ffmpeg smooth this framerate out when creating the output?
[10:26:03 CEST] <JEEB> add the fps video filter I guess
[10:26:07 CEST] <JEEB> if you need a constant frame rate
[10:26:10 CEST] <automatical> I'm finding that the stream is speeding up and the delay from capture to output is getting smaller over time, and then I eventually see buffering when the delay reaches zero
[10:26:35 CEST] <automatical> I'm using that, but I'm not sure if it's detecting the input framerate correctly
[10:26:56 CEST] <JEEB> there's not much to detect, each input frame should have a timestamp
[10:27:05 CEST] <JEEB> of course it could be a random IP camera that has no concept of timestamps
[10:27:11 CEST] <JEEB> in which goot luck and all that crap
[10:27:31 CEST] <automatical> ah :/
[10:27:44 CEST] <automatical> it's a raspberry pi camera
[10:27:54 CEST] <automatical> creating the tcp stream with raspivid
[11:17:12 CEST] <azaki> i'm trying to play around with av1, i encoded a file over several days, and now i'm trying to mux it into an mkv, but it's not working, the file ends up empty. i know the mkv mapping isn't finalized yet, but i figured it'd be using the draft for now or something
[11:17:26 CEST] <azaki> i'm wondering if this is normal for now or if i've run into some bug?
[11:17:55 CEST] <azaki> (this is ffmpeg 4.0.1)
[11:31:40 CEST] <th3_v0ice> How can I remove stream delay using API for example: "Stream #0:0: Audio: aac (LC) (mp4a / 0x6134706D), 48000 Hz, stereo, fltp, delay 1024, 64 kb/s"?
[11:50:54 CEST] <InTheWings> still waiting access to mailing list..
[11:52:39 CEST] <durandal_1707> InTheWings: which mailing list?
[11:52:51 CEST] <InTheWings> ffmpeg-devel
[11:54:00 CEST] <durandal_1707> InTheWings: is there specific reason why you need it?
[11:54:20 CEST] <InTheWings> need to comment on av1 in mp4 patch
[11:55:36 CEST] <durandal_1707> InTheWings: visit #ffmpeg-devel and complain there to llogan or Compn
[12:07:35 CEST] <JEEB> yeh, you'll have to poke people with access to the review list
[13:38:30 CEST] <fsphil> is there any reason not to use the same sws context for scaling two fields in an interlaced frame? it's working for me, but all the examples I've seen use two separate contexts
[13:38:39 CEST] <fsphil> so now I'm worried I've missed something
[13:51:43 CEST] <bencoh> .41
[14:24:04 CEST] <th3_v0ice> While streaming to Twitch from FFmpeg C API twitch shows N/A for video codec. Does anyone know what could be the issue? Outputting to file works like a charm. Muxer is flv, and output is to rtmp.
[14:29:19 CEST] <DHE> fsphil: if the settings for building the sws are identical it should be fine. the sws library should be memoryless (unlike a filter chain)
[14:30:42 CEST] <fsphil> DHE: thanks. yeah the setup is identical for both
[14:30:57 CEST] <DHE> you're not using multithreading are you?
[14:31:20 CEST] <fsphil> not for this bit, both fields are scaled in the same thread one after the other
[14:33:54 CEST] <olspookishmagus> hello, despite using: ffmpeg -i foo.AAC_126kbps_CBR.m4a -codec:a libmp3lame -q:a 6 foo.mp3 , I get a ~= 45.8kbps VBR MP3 ? I was expecting something with higher bit rate.
[14:37:49 CEST] <DHE> olspookishmagus: I believe ffmpeg uses a scale where -q:a 0 is the best quality and 9 is the worst. your setting of 6 is fairly high
[14:37:54 CEST] <DHE> have you tried with a value like 2?
[14:40:42 CEST] <olspookishmagus> I used this a reference: https://trac.ffmpeg.org/wiki/Encode/MP3 which is close to what I have
[14:41:36 CEST] <DHE> is the audio relatively simple? eg: just normal talking vs music with lots of instruments?
[14:52:09 CEST] <olspookishmagus> DHE: it's actually one of the most simple audio clips, a 6-hour long audio clip of white/pink noise that's supposed to be: Spaceship Engine Reactor Noise (6 Hours) [youtube, yvUXAqLakQw]
[15:15:54 CEST] <olspookishmagus> DHE: well that did it, I had to play with the various Q values to get somewhere close to the original bitrate
[15:35:17 CEST] <kepstin> if the audio is mono or there's very little high frequency content, i'd expect lame's vbr mode to use less bitrate, that's pretty normal.
[17:03:03 CEST] <th3_v0ice> Nobody with twitch streaming experience with the API? Or streaming in general.
[18:24:59 CEST] <ayohmang> hi can i ask about configure command here
[18:26:25 CEST] <iive> seems to be the right channel for that
[18:28:55 CEST] <ghoti> totally
[18:30:07 CEST] <BobCat> Don't ask if you can ask, just ask.
[18:30:09 CEST] <ayohmang> im building on solaris 11.3 sparc, configure doesnt work due to sed
[18:32:40 CEST] <ghoti> ayohmang: was there any particular error, or did your computer just vanish into a singularity after you typed the command?
[18:33:05 CEST] <ghoti> ayohmang: also, what version of {everything} are you using?
[18:40:12 CEST] <ayohmang> sed: illegal option -- E , FFMPEG 3.4.git , gcc 4.8.2
[18:41:11 CEST] <ghoti> Righto. The `-E` option for sed works in Linux, macOS, *BSD, etc, but the ancient sed on Solaris apparently doesn't include it.
[18:41:27 CEST] <ghoti> If your package manager allows you to install a more modern alternative, like GNU sed for example, you'd get the option.
[18:42:33 CEST] <ayohmang> i will try compile gnu sed
[18:42:35 CEST] <causasui> I have an opus (audio) file that I'd like to turn into a webm. I don't need any image, just need to play it in a web player that can only play webm video. can ffmpeg do that, and how?
[18:44:52 CEST] <ghoti> ayohmang: probably easier to go with a package manager first. ALso, I think Solaris 11 may already include the GNU sed command, perhaps in `/usr/gnu/bin/sed`.
[18:45:51 CEST] <Shibe> timestamps are unset in a packet of stream 0. what might this mean?
[18:45:54 CEST] <Shibe> that pts is unset?
[18:50:48 CEST] <ayohmang> thank ghoti.
[19:02:32 CEST] <ChocolateArmpits> Shibe, it would seem that dts is unset
[19:02:50 CEST] <ChocolateArmpits> though I don't know where that message originated from
[19:03:50 CEST] <ChocolateArmpits> causasui, if you want to play it extra safe then you can do it like this: ffmpeg -i audio.opus -acodec libvorbis -b:a 128k audio.webm
[19:04:03 CEST] <ChocolateArmpits> or do you need a blank vidoe alongside?
[19:12:17 CEST] <causasui> ChocolateArmpits: I'll try that and see if it works
[19:13:28 CEST] <causasui> ChocolateArmpits: victory, thanks
[19:28:25 CEST] <mont3z> I'm using custom avio to parse a RTP stream. Although I can see that ffmpeg is parsing NAL units at some point av_read_frame blocks for ever. Does anyone knows any reason for av_read_frame to never return the packet?
[19:39:17 CEST] <DHE> causasui: it's possible that "ffmpeg -i audio.opus -c copy audio.webm" will also do the job and be faster if the browser also supports Opus directly
[19:42:27 CEST] <ChocolateArmpits> DHE, more browsers support vorbis though and it's still good enough for most cases
[19:43:18 CEST] <causasui> 4chan still doesn't support opus, just ogg :\
[19:43:21 CEST] <causasui> but it's not 4chan
[19:43:24 CEST] <causasui> so i'm okay :)
[19:48:14 CEST] <iss_> hello, I am trying to recode some data in blender internal process - proxies for videos. This was done by encoding them to MJPEG.
[19:48:54 CEST] <iss_> they lack alpha channel, so I changed the codec, which was OK
[19:49:13 CEST] <iss_> to HUFFYUV
[19:49:32 CEST] <iss_> but I can not change pixel format to encode alpha
[19:50:08 CEST] <iss_> when I look in avcodec_find_encoder(AV_CODEC_ID_HUFFYUV)->pix_fmts there is only AV_PIX_FMT_YUV422P
[19:50:21 CEST] <furq> huffyuv supports bgra
[19:50:30 CEST] <furq> it doesn't support any yuv format with alpha though
[19:51:13 CEST] <kepstin> if you don't need compatibility with apps that don't use ffmpeg, try ffvhuff instead
[19:51:31 CEST] <furq> ffvhuff and ffv1 both support all the yuva formats
[19:51:36 CEST] <iss_> hmm thanks I will try that...
[19:52:00 CEST] <furq> i don't know any other lossless codecs that do yuva
[19:52:40 CEST] <iss_> I don't need lossless, actually lossy format may be better
[19:53:09 CEST] <iss_> I do need indexing, but that is container issue I guess?
[19:53:35 CEST] <kepstin> yeah, just throw whatever you encode into mkv and you should be fine there.
[19:54:27 CEST] <iss_> It worked, great :) thanks a lot!
[19:55:20 CEST] <kepstin> as far as good lossy codecs with alpha, I, uh, really don't know any :/
[19:58:01 CEST] <kepstin> maybe libopenjpeg (a jpeg2000 encoder)?
[19:58:32 CEST] <kepstin> if you were previously using mjpeg, that should behave fairly similarly.
[20:00:51 CEST] <iss_> I mean we can have options -"small" files, but no alpha, or larger files, but alpha...
[20:01:18 CEST] <iss_> if someone wants both they can write a codec :)
[20:13:37 CEST] <furq> kepstin: vp8 does alpha
[20:13:42 CEST] <ghoti> I'd like to take a 1080p video and resize and crop it to play on a 2x2 video wall, where cells are 960x960 pixels. There's a bezel between screens that I've calculated to be approximately 48 pixels (~6.5mm).
[20:13:42 CEST] <furq> maybe vp9 as well
[20:13:48 CEST] <ghoti> I think I can handle the math to take care of cropping out four videos from the original, one for each monitor, then pasting them back together again (hstack/vstack). But is there a better way?
[20:14:11 CEST] <furq> !filter tile @ghoti
[20:14:17 CEST] <furq> uh
[20:14:24 CEST] Action: ghoti holds his breath
[20:14:27 CEST] <ghoti> ;)
[20:14:44 CEST] <furq> !filter tile @ghoti
[20:14:44 CEST] <nfobot> ghoti: http://ffmpeg.org/ffmpeg-filters.html#tile
[20:14:58 CEST] <furq> that takes care of padding for you
[20:15:16 CEST] <furq> oh nvm tile is for successive frames
[20:15:21 CEST] <furq> that probably doesn't work for you
[20:15:56 CEST] <ghoti> Could "successive frames" be expanded to "every frame in the file"?
[20:16:39 CEST] <ghoti> Or .. oh, multiple definitions of "frame" happening here...
[20:17:10 CEST] <ghoti> But thanks, I didn't know about this filter, I'll see if I can learn more in order to ask more specific questions.
[20:25:54 CEST] <ghoti> I guess what I'm thinking I want is to crop out (remove) pixels 937 to 984 in each of X and Y, then either offset everything 24 pixels down+right so that things remain centred.
[20:26:30 CEST] <ghoti> ... or if not offset, then scale everything up, back to 1920x1920, so that the cut line remains at the bezel.
[20:27:21 CEST] <furq> if you just want the stuff occupied by the bezel to be missing then you could just scale
[20:28:15 CEST] <ghoti> Yes, missing is better than having diagonal lines and faces look distorted...  But .. scale what?
[20:28:19 CEST] <furq> or you could just play back the original video, so i'm guessing that's not the answer
[20:28:25 CEST] <furq> it's too hot today
[22:29:41 CEST] <Exagone313> Does ffmpeg supports TAK decoding?  I don't see any compile time option for it in configure --help, but it seems to have been supported at some point (?)
[22:30:22 CEST] <c_14> yes
[22:30:44 CEST] <c_14> it's an internal decoder
[22:30:44 CEST] <furq> Exagone313: builtin codecs aren't listed in configure
[22:30:58 CEST] <furq> they're all enabled by default
[22:30:58 CEST] <c_14> --enable-decoder=tak, but it should be enabled by default
[22:31:03 CEST] <furq> ^ and yes it is
[22:31:43 CEST] <Exagone313> oh, ok, I think it's not what I thought
[22:31:49 CEST] <Exagone313> I get this error from ffplay: TAK codec type 0 is not implemented.
[22:32:11 CEST] <Exagone313> ffmpeg 4.0.1
[22:33:02 CEST] <Exagone313> it's just not implemented I guess. I don't know this format, it does not seem really used and I think the encoder is proprietary (probably patented)
[22:33:57 CEST] <furq> i don't have any tak on hand to test
[22:35:11 CEST] <JEEB> Exagone313: TAK is not patented, just someone thought it was a good idea to make his own thing and of course make it magical closed source stuff
[22:35:29 CEST] <JEEB> also it's just possible that whomever made that decoder didn't have the samples for it
[22:35:42 CEST] <JEEB> if you have a sample, post one on the trac
[22:35:55 CEST] <JEEB> so that if someone cares about TAK codec type 0, they can hack it up
[22:35:55 CEST] <Exagone313> I'm not allowed to have that file so I can't share it
[22:36:36 CEST] <Exagone313> I'll see if I can use the proprietary decoder in a vm or something xD
[22:36:54 CEST] <Exagone313> if it can decode to pcm
[22:37:12 CEST] <Exagone313> thanks anyway!
[22:37:14 CEST] <JEEB> well yea, stuff like foobar2000 etc have support for the decoder AFAIK
[22:37:56 CEST] <JEEB> also if you just cut the sample and post it as a sample media file with a generic file name, nobody really cares :P
[22:38:10 CEST] <JEEB> unless you search trac and there's already a sample posted
[22:38:49 CEST] <JEEB> ok, seems like there's samples already https://trac.ffmpeg.org/ticket/2837
[22:39:06 CEST] <Exagone313> opened 5 years ago :D
[22:39:15 CEST] <durandal_1707> $$$$$$$
[22:39:38 CEST] <JEEB> if you want ¬¬¬¬ kierank already noted what he'd want
[22:39:44 CEST] <Exagone313> I don't have the knowledge to code the thing :/
[22:39:55 CEST] <Exagone313> neither the money to hire
[22:40:08 CEST] <durandal_1707> you have money sure
[22:40:32 CEST] <JEEB> durandal_1707: you're gonna get money out of the broadcast people, not random warez folk :P
[22:40:48 CEST] <JEEB> of course what those two groups people want are different
[22:41:05 CEST] <Exagone313> fuck the guy who thought about using tak instead of flac or wav
[22:41:09 CEST] <durandal_1707> warez folks make cracked software
[22:41:42 CEST] <atomnuker> warez folks have money since they don't buy stuff
[22:42:18 CEST] <Exagone313> what if warez folks are students? :P
[22:42:25 CEST] <furq> i just tried a few different presets and ffmpeg handled them all fine
[22:42:31 CEST] <furq> it's not at all clear from the encoder what type 0 is
[22:42:40 CEST] <durandal_1707> old version
[22:42:42 CEST] <JEEB> furq: with that lol old encoder in that trac issue?
[22:42:48 CEST] <furq> oh i missed that
[22:42:56 CEST] <JEEB> yea modern versions encode newer versions of the format
[22:43:01 CEST] <durandal_1707> only stereo...
[22:43:01 CEST] <furq> fun
[22:43:30 CEST] <furq> i guess all the tak you'll find in the wild is some japanese guy from 2005 who uploaded it to perfect dark and then did a murder-suicide
[22:43:50 CEST] <JEEB> I think at that point it was still share or even winny where it was at
[22:43:53 CEST] <furq> and left us to clean up his mess
[22:43:58 CEST] <durandal_1707> optimfrog
[22:44:07 CEST] <furq> i've never seen anyone use optimfrog
[22:44:09 CEST] <furq> or LA
[22:44:29 CEST] <Exagone313> lol this is japanese music
[22:44:34 CEST] <furq> of course it is
[22:44:56 CEST] <furq> if it's tak, tta, or wavpack, it's from japan
[22:45:00 CEST] <furq> if it's ape then it's either from japan or russia
[22:45:26 CEST] <furq> and if it's shn then it's a grateful dead soundboard from the internet archive
[22:45:34 CEST] <atomnuker> :)
[22:45:44 CEST] <durandal_1707> alac?
[22:45:51 CEST] <furq> i try not to think about alac
[22:46:03 CEST] <JEEB> ALAC at least got fully opened if nothing else
[22:46:17 CEST] <JEEB> as in, the reference encoder/decoder set
[22:46:22 CEST] <Exagone313> yes I did see an ape from russia too
[22:46:41 CEST] <JEEB> also there's at least one store where you can pick between AAC, FLAC and ALAC
[22:47:02 CEST] Action: JEEB bought some music from that place, too bad lossless = audiophile mostly :<
[22:47:11 CEST] <durandal_1707> you can pick aiff
[22:47:26 CEST] <JEEB> and I'm just anal about not buying non-physical that's worse than than the CDs
[22:48:32 CEST] <Exagone313> I'd like to know because either I get physical copies (ok but sometimes higher quality is available, also it's very limited for unknown/foreign artists), either I'm not sure if it's worth buying if you get something worse than physical (like mp3)
[22:49:27 CEST] <durandal_1707> buy lossless, transcode at will
[22:49:44 CEST] <JEEB> yeh
[22:49:58 CEST] <Exagone313> for example, on itunes, you don't know what you get
[22:50:23 CEST] <JEEB> ototoy is nice, lets you pick
[22:50:44 CEST] <JEEB> although it's rather specific
[22:51:04 CEST] <Exagone313> it's not written if you get their alac or aac
[22:52:05 CEST] <Exagone313> there is a french company named qobuz that sell lossless tracks, but they of course don't have unknown artists :P
[22:52:58 CEST] <furq> just be glad it's not tidal mqa
[22:54:18 CEST] <JEEB> oh wow, ototoy finally started adding region limiting. I thought they didn't have it because nothing seemed limited, but now it's just per-item :<
[22:54:32 CEST] <JEEB> which is I guess better than some other stores which are just not letting you buy stuff without a proxy
[00:00:00 CEST] --- Fri Jul 20 2018