[Ffmpeg-devel-irc] ffmpeg-devel.log.20121204

[burek]

[01:25] <teratorn> anyone can point me to example code that synthsizes audio, and encodes it to a container format including pts calculations?
[01:26] <teratorn> I'm really lost on how to set pts values on audio packets when just trying to synthesize a stream of audio data.... it works great in an mp2 file, but a more complicated file using a container format doesn't work - playback sounds bad.
[01:26] <teratorn> err, meant to ask on #ffmpeg :/
[01:39] <cone-746> ffmpeg.git 03Clément BSsch 07b5eb630e216c: lavfi: remove some video w/h settings after avfilter_copy_buffer_ref_props.
[01:54] <cone-746> ffmpeg.git 03Clément BSsch 073313e46c4ae2: lavfi: add subtitles filter.
[02:34] <cone-746> ffmpeg.git 03Clément BSsch 07ef1868351d82: lavfi/removelogo: switch to ff_filter_frame.
[03:49] <cone-746> ffmpeg.git 03Clément BSsch 07359cd12a4324: lavfi/mp: fix invalid read with filters with no argument.
[03:54] <cone-746> ffmpeg.git 03Michael Niedermayer 075e1bacf2d496: matroskadec: reset size when freeing data.
[03:58] <ubitux> michaelni: what's the point of the manual allocation of the buf & bufref in vf mp?
[03:58] <ubitux> when moving to ff_filter_frame, i introduce a leak for some reason
[03:59] <ubitux> i wonder if it's not because the replacement for start_frame within ff_filter_frame is adding a reference
[03:59] <ubitux> (while the previous one was doing nothing)
[04:06] <michaelni> ubitux, how else do you want to allocate them ?
[04:07] <ubitux> well, maybe i'm missing something, but why not make vf mp request a new buffer like any filter, and copy the result of the wrapped filter into that buffer?
[04:08] <ubitux> i guess there are all kind of scenarios where it's not optimal and can not work but well&
[04:08] <ubitux> (btw, divtc has some memory problems)
[04:08] <michaelni> it would need a memcpy of all pixels
[04:08] <ubitux> yeah, not really efficient i guess :)
[04:14] <michaelni> /tmp/ffconf.pgCKE75F.c:2: internal compiler error: in note_invalid_constants, at config/arm/arm.c:11771
[04:15] <ubitux> :(
[04:16] <ubitux> does it happen with l?
[04:17] <michaelni> it happens with a 3 line test from configure
[05:48] <cone-746> ffmpeg.git 03Michael Niedermayer 07400647f9ba6a: configure: Fix ARM thumb detection
[11:13] <cone-982> ffmpeg.git 03Peter Ross 0759fe5a339b75: iff decoder: DEEP TVDC 32-bit decoder
[11:13] <cone-982> ffmpeg.git 03Peter Ross 071696c72af4a5: iff demuxer: include DEEP TVDC lookup table in extradata buffer
[11:30] <saste> ^ spam
[11:30] <mateo`> haha
[11:30] <saste> mateo`, french spam
[11:53] <mateo`> moar french drugs !
[12:01] <saste> more drugs
[12:30] <cone-982> ffmpeg.git 03James Almer 077f0e7fc9e2e6: ast: Rename ast.c -> astdec.c
[12:30] <cone-982> ffmpeg.git 03James Almer 07b70251c0adcc: AST Muxer
[13:00] <Tjoppen> rawvideo seeking is broken
[13:01] <durandal_1707> regression?
[13:01] <Tjoppen> not sure. clicking around in ffplay it often loses the vertical position
[13:02] <Tjoppen> happens in master at least
[13:02] <cone-982> ffmpeg.git 03Paul B Mahol 07266e88a47cfc: mmf: return meaningful error codes
[13:07] <Tjoppen> I'll create a ticket
[13:20] <durandal_1707> yes it must be regression
[13:33] <saste_> ffplay -f lavfi "nullsrc, geq=128 + 107/3*sin(2*(PI/20)*(cos(PI/3)*(X-50*T)+sin(PI/3)*(Y))) + 107/3*sin(2*(PI/20)*(cos(PI/6)*(X-50*T)+sin(PI/6)*(Y)))+ 107/3*sin(2*(PI/20)*(cos(PI/5)*(X*T)+sin(PI/5)*(Y))):128" -noframedrop
[13:36] <Tjoppen> wild plasma
[13:50] <cone-982> ffmpeg.git 03Paul B Mahol 07765debf51f1a: auenc: remove pointless assigment
[13:54] <ubitux> saste_: you can use -f lavfi nullsrc -vf geq=... and no -noframedrop
[13:59] <michaelni> tickets 1987 & 1988 marked as spam and deleted
[14:29] <cone-982> ffmpeg.git 03Diego Biurrun 07201628899b20: fate: ea, h264: prettyprinting and ordering cosmetics
[14:29] <cone-982> ffmpeg.git 03Diego Biurrun 0762641dce7155: build: cosmetics: Move CONFIG_RTPDEC entry to a more suitable place
[14:29] <cone-982> ffmpeg.git 03Luca Barbato 071944d532a8a1: ppc: always use pic for shared libraries
[14:29] <cone-982> ffmpeg.git 03Michael Niedermayer 07dacd6202ec50: Merge remote-tracking branch 'qatar/master'
[14:32] <burek> a little help, if anyone has some spare time..
[14:32] <burek> a guy is trying to remux a video, which is encoded with h.264 inside an .avi container, to a .mp4 container using: ffmpeg -i film.avi -codec copy film.mp4
[14:32] <burek> see the last post here: http://ffmpeg.gusari.org/viewtopic.php?f=11&t=751&p=1287#p1287
[14:32] <burek> there are a lot of errors: [mp4 @ 0x102109800] pts has no value
[14:32] <burek> can anyone figure out what is his issue related to?
[14:33] Action: ubitux recently discovered -fflags +genpts as input option, and wonders if that may help
[14:40] <burek> i'll suggest his input might be broken and tell him to try with +genpts
[14:44] <JEEB> well, AVI has no PTS and only DTS, although the muxed stream might have the information to create the needed information
[14:45] <JEEB> otherwise playback of the file would be impossible, naturally
[14:48] <burek> well, the question is, if that is a bug in ffmpeg or is it a "feature" which could be added in ffmpeg or the input is just broken
[14:49] <Compn> h264 in avi is broken
[14:49] <Compn> to answer your question
[14:50] <JEEB> it's as hacky as MPEG-4 Part 2 in AVI, if not even more >_>
[14:50] <Compn> probably mencoder is the only program that can handle such files :P
[14:50] <JEEB> (I think similar tricks are used to mux both into AVI)
[14:51] <JEEB> also yes, I guess H.264-in-avi support in ffmpeg wasn't/isn't too good :)
[14:52] <Compn> imo whenever 'pts' errors come up, genpts should be autoinserted
[14:54] <iive> pts or dts in avi is not really meaningful, as they are fixed fps (afaik)
[14:54] <burek> ok, now i know what to answer to that guy
[14:54] <burek> thanks a lot for the clarification :)
[14:55] <michaelni> iive, i have variable frame duration avis somewhere
[14:55] <iive> the only way to do vfr is by inserting 0 frames that should repeat the previous frame... now with b-frames this becomes tricky.
[14:55] <michaelni> yes
[15:13] <durandal_1707> michaelni: see #1990, max blocks is 10, how is then 444 done?
[15:23] <michaelni> durandal_1707, smaller macroblocks
[15:24] <michaelni> currently they are fixed at 16x16
[15:24] <michaelni> for 4:4:4 its possible to use 8x8 macroblocks
[15:28] <michaelni> alternatively it could be coded with multiple scans
[15:29] <durandal_1707> michaelni: where are 16x16 hardcoded?
[15:30] <michaelni> at many places, let me think for a moment
[15:35] <michaelni> durandal_1707, i dont see a very easy way to work around that 10 block limit, either the use of mpegvideo (which is quite fixed on16x16) is avoided and dct & quant is called from mjpegenc, or you could try to use 8x16 macroblocks
[15:35] <michaelni> with 8x16 macroblocks you have 6 blocks for 444 
[15:35] <michaelni> and they could be 16x16 on mpegvideo side while mjpeg just stores each as 2 8x16
[15:38] <michaelni> yet another alternative would be to buffer the coeffs from mpegvideo for a whole row and then write them out as 8x8 mbs or as multiple scans of 16x16
[15:50] <durandal_1707> where is 8x16 size for macroblock set?
[15:51] <iive> i'm not quite sure I know what are you talking about. imho when you talk about blocks, they are always 8x8 blocks.
[15:53] <cone-982> ffmpeg.git 03Carl Eugen Hoyos 070877f64fea11: Add debug output when skipping unknown swf tags.
[15:55] <michaelni> iive, blocks are 8x8 but mbs can be various sizes in jpeg
[15:55] <durandal_1707> whatever i hack i get: sampling factors too large for interleaved scan
[15:57] <michaelni> mjpeg_h/vsample in mpegvideo_enc have to be changed
[15:58] <michaelni> for 8x8 4:4:4 mbs they would be all 1 for 8x16 the v ones would stay at 2
[16:00] <michaelni> this is just writing the values in the headers, the actually writen blocks of course need to be changed accordingly
[16:00] <durandal_1707> and how much lines of code needs change?
[16:01] <michaelni> with 8x16 mbs its probably not so much
[16:10] <durandal_1707> and where is other code?
[16:20] <michaelni> the ff_mjpeg_encode_mb() at least, dunno if theres more
[16:20] <michaelni> s/the//
[16:24] <cone-982> ffmpeg.git 03Paul B Mahol 0718aaed6475b0: gifdec: do not handle timestamps
[16:31] <durandal_1707> michaelni: why seeking with mouse for rawvideo demuxer use only byte seek?
[16:32] <durandal_1707> in ffplay
[16:33] <ubitux> maybe because the duration is unknown?
[16:33] <ubitux> so you can't make a pts ratio?
[16:33] <ubitux> (while is size is known)
[16:34] <durandal_1707> ubitux: thanks
[16:34] <ubitux> just a guess heh, i don't know :p
[16:36] <ubitux> with raw the duration should be guessable though
[16:36] <cone-982> ffmpeg.git 03Thierry Foucu 07e98fab1372eb: Update video stats log during encode flush
[16:37] <cone-982> ffmpeg.git 03Michael Niedermayer 0760c2cddff023: ffmpeg: use the correct variables in do_video_stats()
[16:40] <cone-982> ffmpeg.git 03Clément BSsch 073b1b8f88b73e: ffmpeg: small indent fix.
[16:57] <cone-982> ffmpeg.git 03Paul B Mahol 07481c843a4767: rawvideodec: set bit rate
[17:55] <durandal_1707> av_register_protocol is gone]
[18:03] <cone-982> ffmpeg.git 03Paul B Mahol 07a3bb59b163a5: av_register_protocol is gone :(
[18:05] <cone-982> ffmpeg.git 03Michael Niedermayer 0783ddedbbf281: lavf: the core rfps code needs 2 durations to estimate rfps
[19:01] <llogan> any bets on how many more solaris 10 questions we'll get today on ffmpeg-user ML?
[19:02] <llogan> probably dudes from the same office in Mumbai
[19:16] <cone-982> ffmpeg.git 03Nicolas George 070444733b78c0: lavfi/drawtext: add the reload option.
[19:23] <Daemon404> llogan, questions?
[19:23] <Daemon404> libav has a local solaris expert, and i know a bit myself...
[19:42] <llogan> Daemon404: http://ffmpeg.org/pipermail/ffmpeg-user/2012-December/thread.html
[19:43] <llogan> just search "Solaris 10"
[19:43] <JEEB> lol
[19:45] <Daemon404> oh
[19:45] <Daemon404> most of them arent even questions
[19:45] <Daemon404> just "HALP"
[19:45] <llogan> i seem to get more user questions to the mailing list moderator email address from India and neighbors than any other country.
[19:46] <Daemon404> ... i get indians mailing me about canopus lossless
[19:46] <JEEB> lol
[19:46] <llogan> Daemon404: yeah. didn't help that Mr. Gupta keeps making new threads.
[19:46] <JEEB> I think MP4_maniac gets stuff too
[20:22] <burek> i know it's a stupid question in the first place, but still i have to ask it.. is it possible somehow, using ffmpeg, to use -vcodec copy (ip camera input) and to somehow reduce the output fps (from 60 to 30 or similar)
[20:23] <thegeek> perhaps use framestep filter to drop frames?
[20:23] <thegeek> oh wait
[20:23] <thegeek> ip camera
[20:24] <thegeek> what codec?
[20:24] <burek> the point is in using -vcodec copy
[20:24] <thegeek> the machine vision cameras I play with at work are all pretty much simple bitmap formats/codecs or raw bayer
[20:24] <burek> no re-encoding
[20:24] <burek> mpeg4
[20:24] <thegeek> ah ok
[20:24] <thegeek> then ignore me;P
[20:24] <burek> :)
[20:26] <JEEB> in theory if it's all intra you could just leave out Xnth frame and mux those for output and so forth with a lower frame rate. That said, I don't think ffmpeg can do that
[20:27] <JEEB> if it's not all-intra, then nope.avi
[20:28] <nevcairiel> well you could possibyl drop non-ref frames
[20:30] <nevcairiel> no guarantees you'll find many
[20:30] <nevcairiel> on h264 where b-frames are references too, and such :p
[20:34] <burek> well, ok, i figured there is not much ffmpeg can do there, but still was hoping maybe it has got its own magic somewhere inside :)
[20:36] <burek> thanks btw :)
[21:18] <cone-982> ffmpeg.git 03Nicolas George 0703847eb82592: lavf: compute probe buffer size more reliably.
[23:33] <ubitux> saste_: "(from X to Y)"
[23:33] <ubitux> ?
[23:33] <saste_> ubitux: ??
[23:33] <saste_> ah ok understood
[23:35] <ubitux> or even without the "from"
[23:35] <ubitux> -si                <int>   ..F... set stream index (-1 to INT_MAX)
[23:36] <ubitux> any comment on the gradfun & hqdn3d fate tests btw?
[23:38] <ubitux> michaelni: any comment on "[PATCH] lavf: split packets before muxing."?
[23:48] <cbsrobot> ubitux: I get a Program received signal EXC_BAD_ACCESS using -vf subtitles=
[23:49] <cbsrobot> backtrace is
[23:50] <cbsrobot> #0  av_opt_next [inlined] () at /Users/cbsrobot/devel/ffmpeg/libavutil/opt.c:64
[23:50] <cbsrobot> #1  0x00000001006568d1 in av_opt_free (obj=0x102080e00) at opt.c:1068
[23:50] <cbsrobot> #2  0xe000000000000000 in ?? ()
[23:51] <ubitux> mmh
[23:51] <ubitux> cbsrobot: give me a moment, will look at it
[23:52] <cbsrobot> not much help that backtrace - is it ?
[23:53] <ubitux> it's ok, valgrind is complaining
[23:53] <cbsrobot> hehe
[23:54] <cbsrobot> but does subtitles have options ?
[23:54] <cbsrobot> are these the same as -vf ass ?
[23:56] <cone-982> ffmpeg.git 03Clément BSsch 07ab5497df1556: lavfi/subtitles: 10l close codec before format.
[23:56] <ubitux> cbsrobot: can you retry?
[23:56] <ubitux> yes, look at ffmpeg -help full
[23:56] <ubitux> doc is also updated
[23:59] Action: cbsrobot scrolls and scrolls and scrolls
[23:59] <ubitux> less is your friend
[00:00] --- Wed Dec  5 2012