[Ffmpeg-devel-irc] ffmpeg-devel.log.20130424

[burek]

[00:30] <ubitux> yepee i almost have my 2d fft api with ffmpeg
[00:30] <ubitux> i guess i'll be able to write some cool filters soon
[02:09] <cone-623> ffmpeg.git 03Michael Niedermayer 07master:fc6903337188: avcodec/x86/sbrdsp_init: disable using the noise code in x86_64 MSVC, Try #2
[03:02] <ubitux> advise number #1: never ever use only the low 8-bits of a prng to generate a data stream
[03:03] <ubitux> this shit is cycling relatively fast, and thus was fucking my fft results... and of course i thought it was my fft code, while it was just a redundancy in the random i wasn't able to notice at first glance
[03:03] <ubitux> best bug ever.
[03:04] <ubitux> (i had zero bands in my fft results, ’ dafuck)
[03:14] <ubitux> anyway, here is my naive implementation: https://github.com/ubitux/FFmpeg/compare/fft2d-experiments#L2R75
[03:14] <ubitux> the branch include the exact fftw equivalent, results are the same, tested with mismatching w/h
[03:15] Action: ubitux can now start writing funny filters
[03:57] <Compn> ubitux : bright flash remover filter  ? 
[03:57] <Compn> for the poor epileptics
[04:33] <highgod> Hi, I want to ask a question, is there any filters in ffmpeg can tailor the edge of video
[10:02] <cone-797> ffmpeg.git 03Nicolas Bertrand 07master:2e2d24667a2c: jpeg2000: Add mutlti-threading support to decoder
[10:02] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:03b81df9e673: Merge commit '2e2d24667a2c5cc7af94cfa7b63bb9ec1c04224b'
[10:12] <cone-797> ffmpeg.git 03Martin Storsjö 07master:6a8561dbd7c0: x86: Factorize duplicated inline assembly snippets
[10:12] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:c2a0833c0962: Merge commit '6a8561dbd7c078eb75985f7011ad1ad3fda9e223'
[10:17] <cone-797> ffmpeg.git 03Hendrik Leppkes 07master:feec9349d35b: mpegvideo: unref cur/next/prev frames when flushing
[10:17] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:8d86cf4cfa2e: Merge commit 'feec9349d35b3a46d0c6a05e3b23626050b76a77'
[10:24] <cone-797> ffmpeg.git 03Martin Storsjö 07master:486f76f0298a: x86: Get rid of duplication between *_rnd_template.c
[10:24] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:2e789d165b2b: Merge remote-tracking branch 'qatar/master'
[10:34] <ubitux> highgod: vf crop?
[10:34] <ubitux> unless you mean something else?
[10:34] <ubitux> Compn: nope
[10:36] <highgod> thanks
[10:37] <highgod> I will try it
[10:47] <highgod> @ubitux:yes, this filter, thanks
[11:02] <highgod> Hi, I want to ask a quesion, can I use ffprobe to get how long time is the video
[11:22] <ubitux> highgod: -show_format
[11:23] <ubitux> highgod: and you can play with the different ffprobe output (xml, csv, json, ...)
[11:24] <ubitux> you can also just select the field 
[11:36] <highgod> OK, thanks
[12:58] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:62d944539683: vc1: simplify code use INIT_LUT()
[13:04] <ubitux> cvslog dead?
[13:05] <ubitux> j-b: it's not because Skyler_ tried to mailbomb everyone yesterday with x264 that you had to disable ffmpeg cvslog :(
[13:10] <ubitux> oh, finally sent :)
[14:07] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:1d0f817b17d7: vc1dec: Fix mv_f shuffling
[14:07] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:3fdd0979acd3: vc1dec: Fix mv_f shuffling
[14:10] Action: michaelni hides in the corner for failing to realize the 2 commits wherent stashed together as they where supposed to
[14:12] Action: av500 hands michaelni the Narrenkappe
[14:13] <durandal_1707> michaelni: do your work in repo and ask me to merge it when its finished
[14:13] <ubitux> :D
[14:16] <michaelni> durandal_1707, iam happy to send my changes to the vc1 maintainer for merging in his tree, now if only that wasnt defacto myself :(
[14:55] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:72e5d9190843: vc1dec: factorize picture pointer selection code
[15:08] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:236b0c408426: vc1dec: factorize picture pointer selection in vc1_mc_4mv_chroma()
[15:08] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:c5669f3c5996: vc1dec: fix current ptr selection in vc1_mc_4mv_chroma()
[15:27] <ubitux> http://b.pkh.me/lenafft.png
[15:27] <ubitux> i have no idea what i'm doing @_@
[15:28] <durandal_1707> whats that supposed to do?
[15:29] <ubitux> i'm just trying to filter the re/im values of each blocks with an eval
[15:29] <ubitux> here it's "fft='val*lt(hypot(x,y),0.1)'"
[15:29] <ubitux> most likely there is a bug somewhere anyway
[15:30] <ubitux> (x and y being block x/y position in [0,1] scale)
[15:30] <ubitux> i just pushed the filter in my fft2d-experiments branch experiment, if you're interested
[15:34] <ubitux> well maybe that's just because i'm trying to filter them as a circle
[15:36] <michaelni> ubitux, block based transforms lead to blocking artifacts
[15:37] <ubitux> i guess i need some windowing
[15:37] <ubitux> or overlap maybe
[15:37] <michaelni> yes overlapp + window is a possible solution
[15:37] <ubitux> but i believe my eval formula is wrong anyway
[15:37] <ubitux> in several ways :)
[15:37] <durandal_1707> what it shoud do?
[15:37] <michaelni> using a block bigger than your image is another
[15:38] <ubitux> durandal_1707: filter some frequencies :p
[15:38] <ubitux> (drop some)
[15:38] <michaelni> applying the transform with every block position and averaging is yet another but that is slow
[15:39] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:c560437e957c: vc1: make INIT_LUT() self contained
[15:53] <ubitux> BuxiNess: is your mxf sample freely available somewhere?
[15:56] <BuxiNess> yes and no, yes its free but not avlble yet I'll upload it
[15:57] <ubitux> did you try with the threading in ffmpeg if it changes anything?
[15:59] <BuxiNess> Not yet
[15:59] <BuxiNess> ubitux, waht sample do you mean ? The one i'm working in or the fate one?
[16:00] <ubitux> moonrise kingdom sample
[16:00] <ubitux> which you used for benching
[16:00] <kierank> durandal_1707: did you use a specific test sequence to test pullup?
[16:01] <BuxiNess> ubitux, it huge 1.7 Gio.
[16:02] <ubitux> BuxiNess: no worry then
[16:02] <ubitux> i was just curious
[16:04] <durandal_1707> kierank: mp=pullup?
[16:04] <kierank> yes
[16:04] <kierank> i mean did you use a test clip to check the pullup worked
[16:04] <durandal_1707> it does not work for you?
[16:05] <kierank> no, i'm just looking for a hard telecined clip
[16:05] <kierank> to test vivtc with
[16:06] Action: ubitux mainly used http://chromashift.org/kyoani_cm_mechagirl.ts
[16:06] <kierank> thanks
[16:06] <kierank> hopefully that's sfw
[16:06] <ubitux> yes
[16:07] <durandal_1707> kierank: telecine,mp=pullup looks fine
[16:07] <durandal_1707> did not checked other patterns
[16:07] <kierank> oh i was unclear
[16:07] <Compn> hopefully it ISNT
[16:07] <Compn> aww
[16:07] <Compn> cm means commercial
[16:07] <Compn> ... usually
[16:07] <kierank> durandal_1707: i meant i didn't try pullup
[16:07] <kierank> i was just looking for a testclip
[16:07] <kierank> sorry
[16:08] <ubitux> -vf telecine on any clip then
[16:08] <ubitux> :P
[16:08] <kierank> yeah but i wanted a real one
[16:08] <Compn> we should make a set of telecine / telesync / progressive / interlaced clips
[16:09] <durandal_1707> tried blended ones?
[16:10] <nevcairiel> wtf is telesync
[16:10] <Compn> someone who points a camera at a movie screen and records the image with a direct line-in ? 
[16:10] <Compn> :P
[16:11] <ubitux> screener?
[16:11] <Compn> erm records the audio
[16:11] <nevcairiel> seems like the odd man out in your list there :p
[16:12] <JEEB> ubitux, lol you are still using that commercial :D
[16:12] <ubitux> JEEB :D
[16:12] <ubitux> i have a few other telecined sources, but not that much
[16:12] <JEEB> kyoani's commercials are nice, though
[16:12] <ubitux> and i like this one :)
[16:12] <JEEB> yeah
[16:12] <JEEB> there are a couple of others, too
[16:12] <ubitux> feel free to share :)
[16:12] <JEEB> lemme see if I can find them around :)
[16:13] <JEEB> ...if I have the caps any more
[16:13] <JEEB> they usually came in between of kyoani shows
[16:13] <JEEB> (pretty much the only company that doesn't have to sell off its ad space)
[16:15] <ubitux> what is this commercial for btw?
[16:15] <ubitux> a teaser? or just some kind of demo/intro for the company?
[16:16] <ubitux> i btw remember seeing a commercial like this one in nichijou
[16:16] <ubitux> (the wtf one with the dolphins)
[16:17] <JEEB> yeah, it's pretty much an ad for kyoani itself. At some point they made a little note towards kyoani shop (their web shop), but even that commercial was pretty much them demo'ing their skills
[16:17] <JEEB> (one with a flying bus driven by a cat)
[16:19] <JEEB> https://www.youtube.com/watch?v=gEQbgWp2kAs
[16:19] <JEEB> this one
[16:19] <ubitux> heh
[16:20] <ubitux> cute graphics
[16:23] <durandal_1707> cute? thats for kids ....
[16:23] <ubitux> durandal_1707: https://www.youtube.com/watch?v=x7NYrwmlqnU ;)
[16:26] <JEEB> https://www.youtube.com/watch?v=Pel3lW4ZFBQ <- all of kyoani's ads for spring 2012
[16:28] <ubitux> :)
[16:28] <JEEB> and... yeah, it seems like I've removed hyouka, chuu2koi and tamako market caps from my boxes
[16:28] <JEEB> but I bet someone archived those :)
[16:28] <JEEB> because that ts file I cut off my cap for another archiver, too
[16:46] <cone-797> ffmpeg.git 03Nicolas Bertrand 07master:69c02b520bce: MAINTAINERS: add jpeg2000 maintainer
[17:51] <cone-797> ffmpeg.git 03Stefano Sabatini 07master:892af3eb47de: doc/filters: split luma/chroma descriptions for unsharp option
[17:51] <cone-797> ffmpeg.git 03Stefano Sabatini 07master:18144950d46a: doc/filters: introduce example in vflip filter
[17:51] <cone-797> ffmpeg.git 03Stefano Sabatini 07master:6f1716c4b6d4: doc/filters: amend documentation for the hue filter
[17:53] <zimbatm> hi
[17:54] <zimbatm> what tool do you use to debug the structure of an mp4 container ?
[17:54] <zimbatm> apart from the usual ffprobe
[17:54] <zimbatm> i'm trying to figure out what is the content of big headers at the start of a file
[17:55] <zimbatm> that slow down the start of http streaming
[17:55] <nevcairiel> boxdumper (part of l-smash) can give you a detailed dump of the mp4s contents
[17:57] <zimbatm> sweet, thx
[18:34] <Diogo> hi this is a option to loop a video input using ffmpeg?
[18:34] <Diogo> -loop?
[18:34] <durandal_1707> not just vide but audio ...
[18:47] <cone-797> ffmpeg.git 03Georg Martius 07master:4364e1f15097: lavfi: add video stabilization plugins using vid.stab library
[18:47] <cone-797> ffmpeg.git 03Clément BSsch 07master:017051266b37: lavfi/vidstab: miscelenaous cosmetics and typo fixes.
[18:47] <cone-797> ffmpeg.git 03Clément BSsch 07master:cdc48775aa8e: lavfi/vidstab: consistent descriptions.
[18:48] <cone-797> ffmpeg.git 03Clément BSsch 07master:c857c21a8c1c: doc/filters: reword vidstab options intro.
[18:48] <cone-797> ffmpeg.git 03Clément BSsch 07master:9273cc8eb59f: lavfi/vidstab: add missing flags for 2 options.
[18:48] <cone-797> ffmpeg.git 03Clément BSsch 07master:111ac10b60cc: lavfi/vidstab: make use of direct variable.
[18:48] <cone-797> ffmpeg.git 03Clément BSsch 07master:c82f6e58daf6: lavfi/vidstab: remove mention of defaults for consistency with other filters.
[18:48] <cone-797> ffmpeg.git 03Clément BSsch 07master:c7d2a3abf037: lavfi/vidstabdetect: fix direct path.
[19:38] <cone-797> ffmpeg.git 03Clément BSsch 07master:1f6aa7de3ecc: lavfi: add vidstabutils.h to SKIPHEADERS.
[20:01] <saste> ubitux: time to close https://ffmpeg.org/trac/ffmpeg/ticket/1062?
[20:02] <ubitux> oh :)
[20:02] <ubitux> let me do that then
[20:02] <ubitux> delogo has not the flag though
[20:03] <ubitux> let me check if it's possible..
[20:03] <saste> ubitux, that's a separate issue
[20:03] <saste> since this is now generically possible
[20:03] <ubitux> it looks possible
[20:03] <ubitux> there is no frame caching
[20:04] <ubitux> or anything from the previous frames
[20:04] <ubitux> so i'll add that in a moment and close the ticket
[20:04] <saste> ubitux, don't forget removelogo
[20:07] <ubitux> saste: are you sure context is constant in removelogo?
[20:07] <saste> ubitux, i don't know
[20:07] <saste> i think so
[20:10] <ubitux> the direct path in remove logo doesn't make much sense :p
[20:10] <ubitux> it copies all planes anyway
[20:13] <ubitux> it even copies it 2x
[20:14] <cone-797> ffmpeg.git 03Clément BSsch 07master:73050df2402c: lavfi: add timeline support in {de,remove}logo.
[20:14] <nevcairiel> there are two such filters?
[20:16] <ubitux> nevcairiel: one use an external picture as reference
[20:16] <ubitux> while the other well... it just seems to blur the area to fake an interpolation
[20:17] <durandal_1707> what is worst thing ever invented?
[20:18] <ubitux> god
[20:18] <saste> the world
[20:19] <durandal_1707> planar sample format
[20:19] <ubitux> close enough
[20:21] <nevcairiel> planar audio makes sense, if most decoders actually decode to this and then interleave, and a encoder deinterleaves before encoding, being able to skip the step seems like a good idea
[21:54] <kierank> nevcairiel: +1
[21:55] <durandal11707> tell that to users
[21:56] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:782ebd611824: vc1dec: redesign the intensity compensation
[21:56] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:84c0ec92ae79: vc1dec: Implement intensity compensation for vc1_interp_mc()
[21:56] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:100184ccffbf: vc1: use use_ic instead of shuffling mv_mode around
[21:59] <durandal_1707> michaelni: i'm hacking encoder which use put_bits, if put_bits leave uninitialized bytes, ffmpeg encoding becomes very slow and start spamming 0000, enabling assert in put bits does not show anything
[22:00] <durandal_1707> something becomes broken, but i dunno what....
[22:25] <michaelni> i cant guess from that information what the problem might be
[22:31] <tfoucu> I have the coefficient for ebur128 for different sampling rate, like 44.1khz, 32kHz, etc.. looking at the code, right now, the coef are just some define. What will be the right approach to add them? 
[22:32] <durandal_1707> tfoucu: table/switch/whatever
[22:32] <tfoucu> no preference? cool.. I will look at it then.
[22:33] <nevcairiel> should probably talk to the author of that filter, which would be ubitux
[22:33] <tfoucu> that's why i came to IRC to see if he was online and has some idea. 
[22:36] <ubitux> tfoucu: iirc there are some algorithm to compute find the coeff for any freq
[22:37] <durandal_1707> michaelni: you can try it, in dcaenc branch
[22:37] <ubitux> there was a paper about that, i can find it back if necessary
[22:38] <ubitux> also, if you're on this, you can certainly merge the two filters
[22:38] <ubitux> i didn't try to optimize this at all; just made a naive implementation of what the specs said
[22:38] <tfoucu> I have the paper and i even have the compliant audio 48kHz stream if you need them.
[22:39] <tfoucu> ubitux: will you prefer the algorithm to compute them for any freq or just some ahrd coded one?
[22:40] <ubitux> i'd prefer the algorithm
[22:40] <durandal_1707> its would be nice if filter would support any sample rate, so resampling isn't needed at all
[22:40] <ubitux> with the two FILTER() merged into one if possible :)
[22:40] <ubitux> durandal_1707: resampling will be needed anyway when i will add the true peaks
[22:41] <tfoucu> ok.. Let me find again the algorithm and see if I can add it to the filter. 
[22:41] <ubitux> cool :)
[22:41] <tfoucu> and do you want the reference audio file for Fate? only 48kHz
[22:42] <ubitux> ebur128 is covered by fate afaik, with one of the reference 48kHz file
[22:42] <tfoucu> They are from ITU BS 2217
[22:42] <ubitux> i used one from the ebu set
[22:42] <ubitux> isn't it ok?
[22:43] <tfoucu> don;t you need the 5.1 one as well? 
[22:44] <ubitux> yeah well i picked a small and simple one to not have too big samples
[22:44] <ubitux> is it problematic in practice for testing this?
[22:46] <tfoucu> the 5.1 sample from the set is 16MB. I guess it is too big for testing. 
[22:46] <ubitux> even in flac?
[22:47] <tfoucu> the one i have is PCM. Let me encode it to FLAC
[22:47] <ubitux> the current sample is 4.2M
[22:47] <ubitux> the other one was too big
[22:47] <ubitux> but i believe it covers a relatively good part of the filter
[22:48] <ubitux> eventually, we could take 2-3s from that 5.1 sample if you absolutely want to try some other channel layouts
[22:48] <kierank> hi tfoucu 
[22:49] <kierank> i thought ebur128 was only 48khz
[22:49] <tfoucu> the 5.1 in flac will be 2.2MB. 
[22:49] <ubitux> kierank: yeah the specs is defined for 48khz only
[22:49] <ubitux> kierank: but some guys were able to get compute coeff for other frequencies
[22:50] <tfoucu> it's just that in 5.1, the weight for each channel is not just 1. so having a test for 5.1 will validate the weight per channel is correct. 
[22:50] <ubitux> i guess 2.2MB is ok
[22:51] <tfoucu> ok.. Let me check how i can had the test. 
[22:54] <ubitux> ok got the two links
[22:54] <ubitux> but he left :(
[22:59] <kierank> everything should be resampled to a multiple of 48khz :)
[23:00] <kierank> saves a lot of trouble syncing with ntsc
[23:01] <durandal_1707> 6khz
[23:04] <cone-797> ffmpeg.git 03Michael Niedermayer 07master:12e3c1bee487: vc1dec: drop old use_ic code from vc1_b_mc
[23:08] <durandal_1707> michaelni: you compare with reference code or out of head?
[23:19] <michaelni> durandal_1707, the code looked wrong, i did look at the reference sw & spec many times today but not for this one
[23:20] <durandal_1707> michaelni: i'm just interested in high quality decoder
[23:20] <cehoyos> As opposed to what kind of decoder ?
[23:22] <durandal_1707> cehoyos: dirac/j2k/... and incomplete ones (much longer list)
[23:28] <cehoyos> I believe the dirac decoder works pretty well (at least I don't know of any bug reports atm.
[23:29] <durandal_1707> its not bitexact with ref
[23:32] <durandal_1707> cehoyos: nooooo
[23:33] <cehoyos> durandal_1707: You could add a "inclusion of" or similar to the commit message (I misunderstood originally)
[23:34] <durandal_1707> hmm, fps filter should be auto insterted
[23:34] <cehoyos> The relevant commit was not merged
[23:35] <cehoyos> Maybe 74b961d
[23:37] <saste> durandal_1707,  in noise docs: q   higher quality (slightly better looking, slightly slower)
[23:37] <saste> but it is not implemented
[23:37] <durandal_1707> cehoyos: it just replaces, instead of autoinsert for muxers when needed
[23:38] <durandal_1707> saste: so, i'm ignorant, remove it, or add flag to code that does nothing
[23:38] <saste> durandal_1707, was it ever implemented?
[23:38] <durandal_1707> yes, git log is your friend
[23:38] Action: durandal_1707 afk
[23:40] <saste> also it's weird, in mplayer it was "h", not "q"
[23:52] <saste> why all uses of av_dynarray_add() are not checked?
[23:58] <saste> indeed it will crash badly in case of realloc failure, am I right?
[23:58] <saste> so the question is, should we fix it?
[23:59] <saste> ^^ michaelni?
[00:00] --- Thu Apr 25 2013