[FFmpeg-devel-irc] IRC log for 2010-04-10

[irc at mansr.com]

[01:20:43] <blez> I'm sending 200kb chunks thru a named pipe to ffmpeg
[01:20:59] <blez> and near 3mb output size it freezes
[01:21:17] <blez> http://codepaste.net/7bmuha
[01:21:25] <blez> is that a bug?
[05:39:34] <kierank> Tjoppen: tourists seem to like the London Eye.
[06:47:04] <kierank> Tjoppen: also the British Museum (Rosetta Stone, elgin marbles) is worth seeing
[08:24:42] <cecil`> When using -mp3lame you need also to use -ar, but it is not checked. I was thinking about a patch that when the option -mp3lame is used, but not the option -ar gives an error message and quits. Anyone an idea how and where to do this?
[08:25:10] <kshishkov> sounds like a wrong approach at all
[08:26:34] <cecil`> Okay, what would be a good approach?
[08:26:56] <kshishkov> understand what you're doing wrong and don't do it
[08:29:34] <cecil`> That is not very helpful I think. If you do not know that when using -mp3lame, you also need to use -ar (and the documentation does not say it, it even has a example without -ar), then you cann lose valuable time for nothing. What is wrong with giving a message when there is a missing parameter?
[08:33:04] <kshishkov> because you can set -ab instead
[08:34:04] <kshishkov> and not specifying any parameter means you don't know what you're doing
[08:34:10] <jai> you shouldn't _need_ to add -ar
[08:34:34] <jai> cecil`: ^
[08:35:04] <kshishkov> and -aq can be used instead of those two as well
[08:37:35] <jai> of course that too
[08:38:29] <cecil`> Okay, that I did not know. (I am new to ffmpeg.) But I still think that when you use -mp3lame without using -ab, -ar or -aq you should get a message and ffmpeg should stop. I'll post what happens now. Wait a second.
[08:40:26] <cecil`> I tried to convert a video with:
[08:40:26] <cecil`>     ffmpeg -i input.mov -f avi -vcodec mpeg4 -b 400k -g 150 -bf 2 -acodec libmp3lame -ab 64k output.avi
[08:40:27] <cecil`>  
[08:40:27] <cecil`> At  input I get:
[08:40:30] <cecil`>     Stream #0.1(eng): Audio: pcm_u8, 7875 Hz, 1 channels, u8, 63 kb/s
[08:40:33] <cecil`> and at output I get:
[08:40:35] <cecil`>     Stream #0.1(eng): Audio: libmp3lame, 7875 Hz, 1 channels, s16, 64 kb/s
[08:40:38] <cecil`>  
[08:40:41] <cecil`> So far so good, but I also get:
[08:40:45] <cecil`>     Error while opening encoder for output stream #0.1 - maybe incorrect parameters such as bit_rate, rate, width or height
[08:40:48] <cecil`>  
[08:41:08] <kshishkov> ah, so mp3 does not encode that weird sampling rate
[08:41:30] <cecil`> Someone told me to use -ar, that is why I switched to:
[08:41:34] <cecil`>     ffmpeg -i input.mov -f avi -vcodec mpeg4 -b 400k -g 150 -bf 2 -acodec libmp3lame -ar 8000 -ab 64k output.avi
[08:44:41] <cecil`> Yes when using -acodec libmp3lame (not -mp3lame, my bad), you need to give at least one other parameter. When you do not know this, you can spend a 'little' time to get things working.
[08:45:32] <kshishkov> nope, had your file been 8kHz you could get away without -ar *and* -ab
[08:46:34] <kshishkov> and it depends on the library - you go convince libmp3lame devs to print what's wrong on encoding
[08:47:08] <cecil`> Okay, that I did not know. How can I find out what sampling rate my input file uses?
[08:47:36] <kshishkov> it prints that,  use ffprobe or ffmpeg -i infile
[08:48:42] <cecil`> Okay, I was barking up the wrong tree. The problem lies with mp3lame, not with ffmpeg. From your reply I understand there is not much chance they will do about it.
[08:52:39] <cecil`> When using ffmpeg -i, I get:
[08:53:04] <cecil`>     Stream #0.1(eng): Audio: pcm_u8, 7875 Hz, 1 channels, u8, 63 kb/s
[08:53:04] <cecil`>  
[08:54:04] <cecil`> So because it is just a little to low (7875 Hz -> 8kHz) it goes wrong?
[08:54:24] <kshishkov> no, MP3 has some fixed set of sampling rates to enccode
[08:55:07] <cecil`> Where do I find those? And I should always use one of those?
[08:55:26] <cecil`> ffmpeg -i also gives:
[08:55:39] <cecil`>     Seems stream 0 codec frame rate differs from container frame rate: 600.00 (600/1) -> 15.00 (15/1)
[08:55:39] <cecil`>  
[08:56:16] <cecil`> Why is that? (I am new to the video stuff.)
[08:56:24] <kshishkov> ignore that
[08:56:51] <cecil`> Okay, it is nothing to worry about?
[08:57:07] <kshishkov> yes and check http://lame.sourceforge.net for docs on MP3
[08:58:11] <cecil`> I'll do that.
[08:58:36] <kshishkov> and in future you'd better direct such questions to #ffmpeg channel
[08:59:54] <cecil`> Sorry. I thought I had to change something to the ffmpeg code.
[09:02:07] <cecil`> So far so good, but would it not be a nice idea to give a message about the allowed sampling rates when you  get:
[09:02:08] <cecil`>      Error while opening encoder for output stream #0.1 - maybe incorrect parameters such as bit_rate, rate, width or height
[09:02:08] <cecil`> or am I asking to much?
[09:02:32] <kshishkov> you are
[09:02:59] <cecil`> Okay, I'll shut up.
[09:03:09] <cecil`> Thanks for the help.
[09:03:43] <kshishkov> no problem
[09:36:06] <Dark_Shikari> oh god!  the latest MSU codec comparison (I have the draft) includes theora
[09:36:27] <kshishkov> and it bets x264?
[09:36:30] <kshishkov> *beats
[09:37:21] <Dark_Shikari> lol no
[09:37:25] <Dark_Shikari> they intentionally have two copies of all the graphs
[09:37:27] <Dark_Shikari> one set with theora
[09:37:29] <Dark_Shikari> one set without
[09:37:34] <Dark_Shikari> because it screws up the Y axis so badly
[09:37:51] <Dark_Shikari> it gets utterly massacred, way way way worse than xvid
[09:38:02] <Dark_Shikari> I think they must be using an old version, even theora shouldn't be this utterly atrocious
[09:38:15] <kshishkov> while pross-au is not here, what's wrong with xvid?
[09:38:43] <Dark_Shikari> x264 vs xvid is a smaller difference than xvid vs theora.
[09:38:49] <Dark_Shikari> that shouldn't be the case.
[09:39:35] <iive> some new comparison?
[09:41:30] <Dark_Shikari> I have the draft of MSU's latest comparison
[09:41:36] <Dark_Shikari> summary: x264 owns the shit out of everyone by huge margins
[09:41:38] <Dark_Shikari> and they included theora
[09:41:48] <Dark_Shikari> which is so hilariously bad it skews every single graph so much that they include two copies of all the graphs
[09:41:54] <Dark_Shikari> one with theora, one without
[09:42:54] <pJok> in theora, noone can hear you scream
[09:43:27] <Dark_Shikari> it shows x264 as 4 times better than theora
[09:43:37] <Dark_Shikari> on average over their "movie" tests
[09:43:41] <Dark_Shikari> for the same SSIM
[09:44:41] <Dark_Shikari> http://i40.tinypic.com/mslond.png
[09:46:13] <twnqx> lol pJok :P
[09:46:52] <pJok> twnqx, its true... its a video codec... ;)
[09:47:21] <pJok> in vorbis on the other can, anyone with a suitable decoder can hear you scream ;)
[09:48:00] <twnqx> i prefer your screams in alac!
[09:48:14] <twnqx> (just because that doesn't fit in ogg)
[09:48:30] <pJok> hand*
[09:48:54] <iive> pJok: in theora you have no mouth.
[09:49:11] <pJok> i thought that was in the matrix
[09:49:16] <pJok> at least i have no ear
[09:49:17] <pJok> http://right.frequency.dk/~jok/ear.jpg
[09:49:49] <iive> i'm refering to another sci-fi work. "I have no mouth and I can't screen" or something that should mean the same.
[09:50:57] <pJok> ah
[09:51:51] <iive> scream
[09:52:29] <iive> blah... my typing skills are in decline.
[09:54:05] <pJok> hehe
[09:54:09] <pJok> this is the internet
[09:54:14] <pJok> who needs typing skills
[09:54:23] <pJok> they are only for typonazis anyways
[09:55:32] <iive> don't mention that word. I got banned and stripped of rang for using it.
[09:55:57] <pJok> hehe
[10:08:28] <Dark_Shikari> http://i41.tinypic.com/somph3.png http://i44.tinypic.com/2hdbmf8.png
[10:11:03] <Dark_Shikari> so, I take back anything I said about theora getting anywhere near xvid =p
[10:21:31] <kshishkov> Dark_Shikari: you just wait and see ho wXiph interprets that data
[13:01:50] <jai> wb saste
[13:02:32] <pJok> afternoon jai :)
[13:03:49] <saste> jai: hi
[13:05:54] <jai> hey there pJok :)
[13:06:00] <DonDiego> Dark_Shikari: is that msu comparison available somewhere?
[13:36:13] <kierank> ETIMEDOUT isn't defined in mingw if you didn't know already
[13:37:33] <kshishkov> no we don't
[13:37:48] <kshishkov> it's like BSD for Windows anyway in terms of lacking API
[13:38:11] <mru> WinBSD
[13:38:36] * kierank considers finally moving to cygwin
[13:38:46] <mru> why not go all the way to linux?
[13:40:04] <kierank> mainly because of random programs that don't work in wine
[13:40:42] <kierank> but I do have a partition anyway
[14:59:32] <kshishkov> lol, http://google-opensource.blogspot.com/2010/04/interesting-times-for-video-on-web.html
[15:00:05] <jai> yeah :)
[15:04:38] <kshishkov> "The complexity of Theora is considerably less than that of many of its peers" - well, unless you talk about H.264 everything else is _less_ complex
[15:14:07] <merbanan> would be fun if we were able to create a faster implementation for arm
[15:14:43] <mru> is that theorarm thing any good?
[15:14:47] <kshishkov> it's hard since MÃ¥ns has shotgun
[15:15:01] <kierank> merbanan: then someone could write a bizarre blog post to go with it
[15:15:26] <merbanan> yes
[15:15:53] <merbanan> more blog posts
[15:17:31] <kshishkov> you can get bulletproof jacket, jetpack (to counter raised floors) and convince MÃ¥ns to write superfast Theora decoder for ARM
[15:17:38] <kshishkov> and encoder while you're at it
[15:17:58] <merbanan> I think just money would do it
[15:19:14] <kshishkov> you don't have enough
[15:19:42] <mru> I get enough money working on sensible things
[15:21:38] <kshishkov> merbanan: alternatively you can ask Luka, he lives in an area known for people who know how to give an offer you cannot refuse
[15:22:41] <merbanan> :) or you can get a horse head
[15:24:31] <kshishkov> anyway, the main problem is that sets of people who can do stuff for ARM and people who want to do Theora decoder do not intersect
[15:25:11] <kshishkov> and I suspect if you educate people from the second set they will lose interest in it too
[15:29:45] <merbanan> money can motivate, maybe
[15:30:38] <kshishkov> maybe
[16:04:16] <mru> right now I'm motivated to speed up dts decoder
[16:04:40] <kshishkov> tack så mycket
[16:04:54] <kshishkov> if you have any questions don't hesitate to ask Benjamin
[16:05:01] <mru> I have some asm
[16:05:06] <mru> just need a new function pointer
[16:05:46] <kshishkov> which you can reuse for MPEG Audio as well ;)
[16:06:18] <mru> need to decide where to put the pointer
[16:06:27] <kshishkov> dsputil.h
[16:06:52] <mru> we're trying to split that up
[16:07:00] <mru> and it needs fft
[16:07:04] <kshishkov> dsputil_audio.h then ;)
[16:07:10] <mru> nothing in dsputil needs fft now
[16:08:32] <kshishkov> to mdct funcs then
[16:26:28] <merbanan> mru: ok ok
[16:27:12] <mru> wow, that was quick
[16:28:03] <ramiro> mru: in path 2/3, isn't ff_synth_filter_init_arm being called before it's added in patch 3/3?
[16:28:31] <mru> ouch, split them wrongly...
[16:28:34] <mru> oh well
[16:28:41] <CIA-81> ffmpeg: mru * r22826 /trunk/libavcodec/ (dcadata.h dca.c):
[16:28:41] <CIA-81> ffmpeg: DCA: align some arrays
[16:28:41] <CIA-81> ffmpeg: Optimised implementations of the synth filter will require these
[16:28:41] <CIA-81> ffmpeg: arrays 16-byte aligned.
[16:28:41] <CIA-81> ffmpeg: mru * r22827 /trunk/libavcodec/ (dca.c synth_filter.c synth_filter.h): Make synth_filter a function pointer
[16:28:42] <CIA-81> ffmpeg: mru * r22828 /trunk/libavcodec/ (4 files in 2 dirs):
[16:28:42] <CIA-81> ffmpeg: ARM: NEON optimised synth_filter_float
[16:28:42] <CIA-81> ffmpeg: 2.7x faster DCA decoding on Cortex-A8
[16:28:44] <kshishkov> mru: still patch ok
[16:29:57] <mru> next: try to speed it up another 30%
[16:30:47] <kshishkov> patch ok :)
[16:31:39] <kshishkov> DCA decoder has two maintainers and both are quite permissive ;)
[16:58:37] <_av500_> \\\ooo///
[17:01:38] * DonDiego fears that _av500_ morphed into a weird three-headed alien..
[17:03:14] <mru> no, that troll from the holy grail
[17:14:36] <kshishkov> mru: did you mean three-headed knight or the monster from castle Aargh?
[17:16:59] <mru> I mean the 3-headed thing sir robin ran away from
[17:17:28] <kshishkov> a knight
[17:17:42] <kshishkov> s/ran away/bravely ran away/
[17:17:57] <mru> that was no knight
[17:18:24] <_av500_> calm down it is only motion blur
[17:18:57] <mru> "it's only a model"
[17:19:18] <kshishkov> http://www.youtube.com/watch?v=c4SJ0xR2_bQ
[17:33:36] <CIA-81> ffmpeg: jai_menon * r22829 /trunk/libavcodec/alacenc.c: alacenc : Report supported input sample formats.
[18:08:39] <kshishkov> .quit
[19:01:16] <Dark_Shikari> DonDiego: not published yet
[19:05:19] <Dark_Shikari> siretart: ping
[19:13:13] <DonDiego> Dark_Shikari: when will it get published?
[19:14:10] <Dark_Shikari> I'll upload it somewhere.
[19:16:07] <iive> encoders comparison? i hope there is latest theora beta/dev version too, or they will say "you used older theora encoders on purpose" :)
[19:17:13] <DonDiego> yes
[19:20:49] <Dark_Shikari> it's the last theora release :)
[19:21:48] <Dark_Shikari> http://www.mediafire.com/?djym2vmjddm <--comparison
[19:22:01] <iive> well, then you should compare it to the last x264 release :P
[19:22:21] * iive tries to remember have there ever been x264 release.
[19:22:43] <Dark_Shikari> This is why we don't have releases!
[19:22:46] <Dark_Shikari> I've said this over and over
[19:22:56] <Dark_Shikari> If you don't have releases, nobody can get away with "using the last release" because there is none ;)
[19:24:33] <DonDiego> Dark_Shikari: can i pass this thing around?
[19:25:09] <Dark_Shikari> hmm... sure.
[19:25:13] <DonDiego> ok
[19:25:20] <DonDiego> anyway, gtg, cu guys
[19:29:16] <iive> Dark_Shikari: honestly. Don't give them escape route. If you make the comparison, add the latest dev version, so the people would know that even next release won't be as good as xvid.
[19:29:32] <Dark_Shikari> They'd just say you were using a buggy recent version
[19:29:37] <Dark_Shikari> and you should use the last release
[19:29:43] <iive> keep both.
[19:29:47] <Dark_Shikari> lol
[19:34:19] <iive> if you want just one version, then make the test and include the one that is better. then mention the other performs worse.
[19:39:43] * elenril lols at usage of AndZoidberg in that comparison
[19:43:49] <Dark_Shikari> it's true
[19:52:56] <Compn> so uh
[19:53:18] <Compn> nm, must be mp3 encoding time
[20:32:26] <BBB> so google funds (in some way) theora now?
[20:32:35] <BBB> that's an interesting turn of events... </sarcasm> :)
[20:34:26] * _av500_ remembers the theorarm from some time ago...
[20:34:38] <_av500_> so what is new?
[20:35:30] <BBB> I just heard about that today - I'm surprised, I'd figure they'd release vp8 by now
[20:35:44] <CIA-81> ffmpeg: koorogi * r22830 /trunk/ (libavfilter/avfilter.h Changelog): Bump libavfilter minor version and Changelog for unsharp filter
[20:35:51] <BBB> instead, they're still monkeying around with the 12-year-old vp3-related technology
[20:36:30] <_av500_> ill wait for it to alpear in android..
[20:37:01] <_av500_> if they are serious..
[20:38:08] <Compn> i'm still betting that google sits on all of on2 until h264 charges for web licenses
[20:38:09] <Dark_Shikari> what's the news?
[20:38:21] <Compn> then google can charge less and win the internet video wars
[20:38:46] <_av500_> Compn: cool. so never :)
[20:39:26] <BBB> that's a great way to negotiate
[20:39:38] <BBB> well, good for google with their big walle
[20:39:44] <Dark_Shikari> what's this news today?
[20:40:05] <_av500_> googtheora
[20:40:08] <Dark_Shikari> where
[20:40:41] * _av500_ is in tram cannot paste url
[20:41:08] <_av500_> scroll back
[20:42:07] <Dark_Shikari> I don't see it
[20:43:30] <_av500_> google opensrc blog
[20:43:34] <BBB> http://google-opensource.blogspot.com/2010/04/interesting-times-for-video-on-web.html
[20:44:07] <_av500_> thx
[20:44:53] <Dark_Shikari> lol
[20:44:55] <Dark_Shikari> "stands up quite well"
[20:44:58] <Dark_Shikari> "being simpler to decode"
[20:45:07] <_av500_> :)
[20:45:07] <Dark_Shikari> mpeg-4 sp is lower complexity than theora
[20:45:19] <_av500_> h263 ftw
[20:45:24] <Dark_Shikari> even google has drunk the koolaid now
[20:45:31] <Dark_Shikari> well, odds are more likely it's just the guy writing the blog post
[20:45:31] <_av500_> goolaid
[20:45:56] <_av500_> 100% evil free
[20:46:39] <BBB> #define evil
[20:47:19] <_av500_> !nice
[22:25:41] <_av500_> Dark_Shikari: lol at how they need to remove theora from the graphs
[22:26:10] <_av500_> that is the most impressive argument
[22:50:27] <Dark_Shikari> _av500_: lol yes
[23:44:51] <Kovensky> where's vitor :X
[23:44:58] <Kovensky> (or someone else that knows about audio filtering)