[Ffmpeg-devel-irc] ffmpeg-devel.log.20130913

[burek]

[00:06] <cone-47> ffmpeg.git 03Vignesh Venkatasubramanian 07master:1d0d2201b4f4: Adding a line to Changelog about Opus Matroska Support
[00:06] <cone-47> ffmpeg.git 03Vignesh Venkatasubramanian 07master:b7bd68853d67: lavf/matroska: Removing experimental suffix for Opus
[07:39] <ubitux> wtf...
[10:56] <durandal_1707> michaelni: how you got crash that fixes 51dab60c7b91a21ec280c7d3042aa8350aae048f ?
[10:57] <durandal_1707> i always get same numbers with bytewidth and linesize
[11:18] <ubitux> Daemon404: fyi, vid.stab merged my work for linesize fixes
[11:18] <ubitux> and i fixed the green artefacts
[11:18] <ubitux> so it's usable now
[11:21] <Daemon404> ubitux, "usable"
[11:21] <Daemon404> maybe by hand
[11:22] <ubitux> ?
[11:22] <Daemon404> but it's e.g. cropping is still busted
[11:22] <Daemon404> unless you tweak it
[11:22] <Daemon404> youll see random black borders appear, depending on how unstable it is
[11:22] <Daemon404> it fails to calculate how to much to rop properly
[11:22] <Daemon404> not useful in a pragmatic waya :P
[11:22] <Daemon404> -a
[11:22] <ubitux> probably
[11:47] <cone-915> ffmpeg.git 03Georg Martius 07master:ad96482d67b2: avfilter/vidstabtransform: allow negative zoom.
[12:07] <Daemon404> ubitux, i maybe look into it one day
[12:07] <Daemon404> i mean.. in theory it's possible to calculate exactly how much to crop
[12:07] <Daemon404> given a certain transform/shear
[12:10] <ubitux> i'm working on it currently
[12:10] <Daemon404> oh
[12:10] <Daemon404> well then.
[12:10] Action: Daemon404 goes back to being laz
[12:10] <Daemon404> y
[12:11] <ubitux> it will probably be merged within ffmpeg
[12:12] <ubitux> in an undefined future
[12:15] <Daemon404> ubitux, i hope you also mean to drop 99% of the code
[12:15] <Daemon404> which is all available in e.g. swscale
[12:15] <Daemon404> ;)
[12:15] <ubitux> yeah probably
[12:15] <ubitux> and benefit from our optims...
[12:17] <Daemon404> ubitux, swscale is actually very slow in man instances :P
[12:17] <Daemon404> compaed to how it could be
[12:17] <Daemon404> e.g. extra conversions, etc
[12:17] <durandal_1707> what to drop from swscale?
[12:17] <Daemon404> durandal_1707, it's an architecture problem
[12:17] <Daemon404> not easily fixed
[12:17] <ubitux> Daemon404: vid.stab code is really naive
[12:18] <ubitux> so it will be hardly slower
[12:18] <ubitux> note that the vid.stab performances are particularly bad
[12:18] <ubitux> not*
[12:18] <Daemon404> oh
[12:18] <Daemon404> i know
[12:18] <Daemon404> it's super slow
[12:18] <ubitux> not that much
[12:18] <ubitux> IMO
[12:18] <Daemon404> i mean a copy is bundled with opencv.... kind of slowness
[12:18] <Daemon404> king*
[12:18] <Daemon404> ;)
[12:19] <ubitux> huh? what's this opencv thing about?
[12:20] <Daemon404> ubitux, opencv includes a copy of vid.stab
[12:20] <Daemon404> bundled
[12:20] <Daemon404> is all.
[12:20] <ubitux> wtf
[12:20] <Daemon404> opencv is horribly managed... 
[12:20] <ubitux> ok..
[12:20] <wm4> what's opencv again?
[12:21] <durandal_1707> we have opencv wrapper, so lets dicht both filters
[12:21] <Daemon404> used for stuff liek object tracking
[12:21] <Daemon404> facial recog
[12:21] <Daemon404> etc
[12:21] <nevcairiel> open library of video algorithms
[12:21] <nevcairiel> rather generic collection of everything :p
[12:21] <Daemon404> e.g. recognize a banana in a photo
[12:21] <wm4> sounds useful
[12:21] <Daemon404> it is
[12:21] <Daemon404> but to do so you have to train it
[12:21] <Daemon404> with 100s or 1000s of photos of bananas
[12:21] <Daemon404> :D
[12:22] <wm4> I don't have that many bananas
[12:22] <durandal_1707> lool
[12:22] <wm4> speaking of external stuff, how useful is that frei0r stuff?
[12:23] <Daemon404> not useful
[12:23] <wm4> and why doesn't lavfi have wrappers for ladspa and sox?
[12:23] <Daemon404> sox wrapper is in swresample
[12:23] <Daemon404> iric.
[12:23] <Daemon404> er, iirc.
[12:23] <wm4> only for resampling, isn't it
[12:23] <Daemon404> yeah
[12:24] <Daemon404> there was a vf_sox.c on the ML from.. stefano
[12:24] <Daemon404> i think
[12:24] <Daemon404> one time.
[12:24] <durandal11707> lavfi does not have ladspa wrappers because nobody wrote it
[12:25] <wm4> I mean, after all mplayer has a ladpsa wrapper
[12:25] <wm4> and isn't all mplayer filter code considered holy to lavfi devs?
[12:26] <ubitux> 12 filters left
[12:26] <Daemon404> wm4, holy?
[12:26] <Daemon404> nah
[12:26] <ubitux> well, 12 .c
[12:26] <ubitux> less then
[12:26] <Daemon404> perhaps "hole-y"
[12:26] <Daemon404> :)
[12:26] <durandal11707> the problem is only libmpcodecs is considered holy, af2 not 
[12:26] <durandal11707> libaf2 was never merged, ask michaelni why 
[12:27] <durandal11707> ah, perhaps because lavfi did not have audio support then?!
[12:27] <wm4> don't give him bad ideas
[12:28] <ubitux> :D
[12:45] <wm4> why does AVPicture exist?
[12:45] <wm4> it duplicates AVFrame stuff
[12:45] <wm4> historic reasons?
[12:47] <durandal11707> the big question is what from libaf2 is considered worthy enough to be ported/reimplemented?
[12:49] <wm4> not much
[12:49] <wm4> it's a big loads of crap, basically
[12:50] Action: wm4 looks at af_sweep
[12:56] <durandal11707> ubitux: add contrast and gamma to hue and it will be 2 less
[12:57] <durandal11707> perhaps i should finally write adelay filter?
[12:57] <durandal11707> and port pullup
[12:59] <durandal11707> but i want to make transpose 6x faster
[13:10] <BoR0> I have a C++ library that wraps around libffplay, and instead of doing while (position != y...) fetch audio from ffmpeg I need to fill packets to directx as soon as ffplay has data for me
[13:10] <BoR0> does anyone have any experience with DirectSound? basically I need a queue management with audio buffer rather than setting notifications on position and filling buffer accordingly. is it possible to do this with plain DirectSound API or does there exist any nice wrapper for this?
[13:16] <wm4> libffplay? what
[13:19] <BoR0> yes, I have converted ffplay to a libffplay
[13:19] <BoR0> and exposing methods that will give me raw video and audio data
[13:28] <durandal11707> why this does not work: ffmpeg -f lavfi -i "aevalsrc=0|0|0|0" -lavfi "channelsplit=channel_layout=4.0:[A1][A2][A3][A4];[A1][A2] amerge=inputs=2 [o0];[A3][A4] amerge=inputs=2 [o0]" -f null -
[13:29] <wm4> durandal11707: it hates you
[13:42] <durandal11707> anyone?
[13:43] <wm4> well, maybe you should say WHAT "not works" means
[13:43] <wm4> because "it doesn't work" is the worst problem description you could possibly give
[13:46] <durandal11707> it gives me: Output pad "default" with type audio of the filter instance "Parsed_amerge_1" of amerge not connected to any destination
[13:48] <durandal11707> ah, i need to add -map '[o0]' -map '[o1]'
[13:49] <cone-915> ffmpeg.git 03Michael Niedermayer 07master:53102fa2f044: Revert "avfilter/vf_noise: dont corrupt the picture outside width x height"
[13:49] <cone-915> ffmpeg.git 03Michael Niedermayer 07master:126ba688267e: avfilter/vf_noise: rename linesize to bytewidth
[13:49] <cone-915> ffmpeg.git 03Christian Schmidt 07master:d9cdb7d8d6d8: pcm-dvd: Support channel configuration changes
[13:49] <cone-915> ffmpeg.git 03Michael Niedermayer 07master:3c89ec281695: Merge commit 'd9cdb7d8d6d828bb5497ea3f0fd7edd2f3f6cc30'
[13:50] <durandal11707> michaelni: but that is saste ideas to name it linesize and not stride/or bytewidth
[13:51] <durandal11707> have bounties be added to trac?
[13:52] <durandal11707> gmake: *** No rule to make target `doc/avtools-common-opts.texi', needed by `doc/ffmpeg.pod'.  Stop.
[13:53] <cone-915> ffmpeg.git 03Christian Schmidt 07master:aca25104ea9c: pcm-dvd: Minor leftovers
[13:53] <cone-915> ffmpeg.git 03Michael Niedermayer 07master:fc477263fadd: Merge remote-tracking branch 'qatar/master'
[15:10] <durandal11707> hmm channelsplit filter clones input frames, thus it wastes memory
[15:40] <durandal11707> hmm that is correct only if reference counting is disabled
[17:56] <durandal_1707> how to make one filter calls another one ?
[18:02] <wm4> use avisynth
[18:02] <wm4> </troll>
[18:03] <gnafu> Seems legit.
[18:07] <durandal_1707> well on the other hand calling reconfiguration from filter itself
[18:08] <durandal_1707> so that auto filter insertion like scale kicks in
[18:30] <durandal_1707> one more filter under way
[20:38] <llogan> "buttracker"
[20:38] <Daemon404> with opencv?
[20:58] <cone-25> ffmpeg.git 03Michael Niedermayer 07master:a25585bb5072: avutil/pixdesc: Prevent minor array overread in ff_check_pixfmt_descriptors()
[20:58] <cone-25> ffmpeg.git 03Michael Niedermayer 07master:f3ba91a3f1d9: avutil/pixdesc: dont try to use av_read_image_line() with bayer formats
[21:18] <cone-25> ffmpeg.git 03Paul B Mahol 07master:b8b398027cc2: avfilter/video: remove commented out cruft
[21:29] <cone-25> ffmpeg.git 03Paul B Mahol 07master:3dcd3e08f6a4: avcodec/proresdec2: remove duplicated tables
[21:40] <durandal_1707> llogan: how this sounds "If number of given delays is smaller than number of channels all remaining channels will not be delayed."
[21:48] <cbsrobot_>  durandal_1707 and what happens if it's bigger ?
[21:48] <durandal_1707> ignored
[21:49] <cbsrobot_> then add a comma between channels and all
[21:50] <durandal_1707> why?
[21:58] <cbsrobot_> because there are two verbs and two subjects
[21:59] <cbsrobot_> but I'm not native speaker
[22:08] <durandal_1707> what ffv1 doc change i need to update?
[22:09] <llogan> durandal_1707: sounds good to me
[22:22] <durandal_1707> llogan: adding twitter link to main page?
[22:29] <llogan> next to the g+?
[22:31] <durandal_1707> whatever
[22:32] <durandal_1707> all icons could be on right side....
[22:33] <durandal_1707> then you also need to create fb one
[22:34] <durandal_1707> or better not...
[22:37] <llogan> yes, icons would be nicer on right, but i don't really care about the fb one.
[22:37] <llogan> maybe i'll make patches over the weekend
[22:41] <cone-25> ffmpeg.git 03Paul B Mahol 07master:23bd0335af70: avcodec/g2meet: remove redundant log message
[23:30] <cone-25> ffmpeg.git 03Michael Niedermayer 07master:6c07e41f4302: avcodec/mjpegdec: check for malloc failure of blocks[] and last_nnz[]
[00:00] --- Sat Sep 14 2013