[Ffmpeg-devel-irc] ffmpeg-devel.log.20180814

burek burek021 at gmail.com
Wed Aug 15 03:05:03 EEST 2018

[00:28:58 CEST] <TD-Linux> jamrial, sent patch
[01:38:03 CEST] <TD-Linux> jamrial, actually hold a bit before landing my patch, there seems to still be something broken with fmp4
[01:43:18 CEST] <TD-Linux> apparently mov_find_codec_tag bypasses mov_get_codec_tag sometimes? what am I supposed to do here?
[01:51:46 CEST] <TD-Linux> I can force mov_find_codec_tag to use mov_get_codec_tag for the cases I care about
[02:08:48 CEST] <jamrial> TD-Linux: ah, it doesn't call mov_get_codec_tag() for mp4
[02:12:02 CEST] <jamrial> guess the codec_tag check in init_output_stream_streamcopy() in ffmpeg.c needs to be fixed/changed to really solve this
[02:23:32 CEST] <TD-Linux> jamrial, I could also hack the ivf parser. but I don't know what else could potentially produce bonkers tags
[02:24:07 CEST] <TD-Linux> jamrial, another option is to force the MP4 case through mov_get_codec_tag. that seems to work locally
[02:27:46 CEST] <jamrial> yes, but it's ugly since mov_get_codec_tag() is explicitly for mov output
[02:29:00 CEST] <jamrial> you could maybe add a mp4_get_codec_tag(), called from mov_find_codec_tag() when track->mode == MODE_MP4, and add this specific check there
[02:50:48 CEST] <TD-Linux> jamrial, yeah I see there is an array for mp4 tags specifically.
[05:58:51 CEST] <TD-Linux> jamrial, when writing a non-fmp4, I am not getting any trak atoms written. any idea why that might be?
[06:00:20 CEST] <jamrial> TD-Linux: what is "fmp4"?
[06:00:46 CEST] <TD-Linux> jamrial, movflags frag_keyframe
[06:03:18 CEST] <TD-Linux> it looks like tracks[i].entry would be 0 but I'm not sure why that would be
[06:50:18 CEST] <TD-Linux> okay this is only an issue when remuxing from ivf
[06:50:50 CEST] <TD-Linux> and is actually because it never gets any packets from the ivf.
[06:52:14 CEST] <TD-Linux> files made with ffmpeg's libaom codec work totally fine with my mp4 reader. cool.
[09:18:43 CEST] <gagandeepsingh> kierank: do i send the patches again writing patch 1/3, patch 2/3, patch 3/3 ?
[09:19:02 CEST] <gagandeepsingh> damn messed up again on ml :\
[09:20:16 CEST] <kierank> Yes
[09:32:59 CEST] <gagandeepsingh> if i have a patch containing multiple commits, will it send a batch of mails to ml
[09:38:15 CEST] <gagandeepsingh> lesson learned :\
[09:38:45 CEST] <gagandeepsingh> believe me i am not trying to spam mailing list
[09:44:03 CEST] <akravchenko188> jkqxz: Hi, I have sent reminder to forum. did you have a chance to review the patches? thanks
[09:45:23 CEST] <gagandeepsingh> kierank: i have manually attached patches to mails
[09:46:24 CEST] <kierank> Ok
[09:48:19 CEST] <akravchenko188> jkqxz: threads begin in July, should I resend them one more time this month?
[13:56:03 CEST] <gagandeepsingh> kierank: https://docs.google.com/document/d/1sUswaJW5tkGj0nQavwZHG20d_rLKvxGbHgrWBQLiklY/edit?usp=sharing
[13:56:22 CEST] <gagandeepsingh> link i am going to send in evaluations, just give it a read
[13:57:16 CEST] <kierank> ok
[13:57:28 CEST] <kierank> gagandeepsingh: what does mountain sample look like (screenshot) in the end?
[13:59:37 CEST] <kierank> gagandeepsingh: google doc looks fine
[14:05:24 CEST] <gagandeepsingh> kierank: https://imgur.com/a/jX4llAM
[14:05:28 CEST] <gagandeepsingh> screenshots
[14:05:36 CEST] <kierank> nice
[14:05:37 CEST] <gagandeepsingh> also how do i send the 3 samples i have
[14:05:43 CEST] <kierank> ask michaelni 
[14:05:49 CEST] <kierank> michaelni: can you help gagandeepsingh 
[14:47:22 CEST] <gagandeepsingh> kierank: evaluation form filled
[15:37:53 CEST] <Mina_> Hi, I was wondering why doesn't FFmpeg have a mean video filter?
[15:38:11 CEST] <durandal_1707> median?
[15:38:26 CEST] <Mina_> Yes, an averaging filter similar to gauss.
[15:38:30 CEST] <durandal_1707> 3x3 or higher?
[15:38:38 CEST] <Mina_> Any
[15:39:18 CEST] <Mina_> Or does it have one that I'm not aware of?
[15:39:30 CEST] <durandal_1707> 3x3 median is in removegrain mode
[15:40:18 CEST] <durandal_1707> others are not present althought therre is O(1) algo
[15:40:43 CEST] <durandal_1707> avtually MxN
[15:41:15 CEST] <thardin> surely there's a convolution filter?
[15:41:55 CEST] <Mina_> Just checked it, sorry wasn't clear for me it existed. The reason I am asking is that I found a log2(r) version of it and thought of comparing to the one present.
[15:42:19 CEST] <durandal_1707> there is even better version
[15:42:32 CEST] <Mina_> That's the one present?
[15:43:41 CEST] <durandal_1707> no
[15:43:46 CEST] <durandal_1707> https://github.com/HomeOfVapourSynthEvolution/VapourSynth-CTMF
[15:46:26 CEST] <Mina_> License issues?
[15:48:25 CEST] <Mina_> I mean is that the reason why it's not used?
[15:48:40 CEST] <thardin> probably no one caring enough
[15:48:50 CEST] <durandal_1707> its not lgpl
[15:49:04 CEST] <durandal_1707> so i hadnt ported it
[15:49:21 CEST] <Mina_> Thought so, but can we just implement the paper? The paper has no license i assume.
[15:49:34 CEST] <durandal_1707> i guess so
[15:50:10 CEST] <Mina_> Well I'll probably investigate more about it and hopefully do just that.
[15:54:22 CEST] <thardin> clever little algorithm
[15:56:24 CEST] <thardin> very vectorizeable even
[15:57:29 CEST] <Mina_> Haven't looked at it in detail yet but sounds cool.
[15:58:16 CEST] <thardin> first you can update all the column histograms in paralllel. plus the histogram merging can be sped up with some appropriate SSE ops
[15:58:35 CEST] <thardin> GPU implementation should also be easy
[15:59:26 CEST] <thardin> or hm.. maybe not quite as GPU friendly. but still
[16:01:01 CEST] <Mina_> durandal_1707: Thanks anyway.
[16:12:42 CEST] <thardin> it's implemented in opencv. I presume one can access it if building with --enable-gpl
[16:14:02 CEST] <JEEB> IIRC opencv's biggest problem is that they stopped supporting the C API that FFmeg is using
[16:14:19 CEST] <durandal_1707> ocv filter is very limited
[16:14:21 CEST] <thardin> oh yeah, they went to C++ and ABI hell
[16:14:26 CEST] <JEEB> yup
[16:14:49 CEST] <JEEB> https://github.com/opencv/opencv/issues/8438#issuecomment-288638915
[16:17:46 CEST] <thardin> you might want to reimplement it anyway, since https://nomis80.org/ctmf.c only does padding with zeroes
[16:18:06 CEST] <durandal_1707> no AVI experts here?
[16:21:52 CEST] <JEEB> durandal_1707: I thought it was based on RIFF or so?
[16:22:06 CEST] <JEEB> I think MS even gave a relatively "OK" spec on AVI
[16:22:24 CEST] <JEEB> and then there's virtualdub too as a relatively "nice" reference, I guess?
[16:25:18 CEST] <durandal_1707> there is mplayer so its demuxer avi is demuxing all packets
[17:34:15 CEST] <durandal_1707> avi demuxer parses all stuff from index but somehow manages to drop 35 keyframes
[17:38:24 CEST] <durandal_1707> the avi demuxer code is teh ugly
[18:12:09 CEST] <JEEB> durandal_1707: I bet there's various hacks for semi-broken files there?
[18:12:38 CEST] <JEEB> I wouldn't be surprised if such old things would have to be reworked to be more readable/understandable
[18:20:01 CEST] <durandal_1707> unreadable hacking mess
[18:24:23 CEST] <cone-931> ffmpeg 03Sergey Lavrushkin 07master:582bc5a348f5: libswscale: Adds conversions from/to float gray format.
[18:55:15 CEST] <cone-931> ffmpeg 03Gyan Doshi 07master:02741853d74d: avformat: correct fdebug loglevel
[19:01:47 CEST] <cone-931> ffmpeg 03Thilo Borgmann 07master:72d9b8f4c53c: lavc/videotoolboxenc: Fix compilation on osx 10.10.5 Yosemite
[19:02:23 CEST] <cone-931> ffmpeg 03Thilo Borgmann 07release/4.0:49a90d5d3143: lavc/videotoolboxenc: Fix compilation on osx 10.10.5 Yosemite
[20:10:52 CEST] <durandal_1707> i fixed avi bug, I am now the AVI expert
[20:12:30 CEST] <TD-Linux> will you implement av1 in avi
[20:13:33 CEST] <durandal_1707> lol , if you pay enough i will
[20:16:49 CEST] <jamrial> you can probably already mux av1 in avi. libavformat will just mux anything as long as it has a riff tag. there are zero constrains there
[20:16:55 CEST] <jamrial> the output being readable is a whole different story :p
[20:23:51 CEST] <TD-Linux> so I can have both av01 and AV01 there? :)
[20:30:50 CEST] <kierank> durandal_1707: lolwut
[20:30:51 CEST] <kierank> that patch
[20:33:01 CEST] <durandal_1707> what?
[20:37:07 CEST] <cone-931> ffmpeg 03Sergey Lavrushkin 07master:2ecf9d103a73: libavfilter/dnn_backend_tf.c: Fixes ff_dnn_free_model_tf.
[20:37:08 CEST] <cone-931> ffmpeg 03Sergey Lavrushkin 07master:95cb2127adcf: libavfilter/vf_sr.c: Removes uint8 -> float and float -> uint8 conversions.
[20:37:09 CEST] <cone-931> ffmpeg 03Sergey Lavrushkin 07master:1f18a633c601: libavfilter: Adds proper file descriptions to dnn_srcnn.h and dnn_espcn.h.
[20:39:55 CEST] <jamrial> TD-Linux: regarding what you asked yesterday, i couldn't reproduce
[20:40:05 CEST] <jamrial> using that flag i still get trak boxes
[20:40:16 CEST] <TD-Linux> jamrial, it turns out you only don't get them with ivf files
[20:40:40 CEST] <TD-Linux> it's some bug in ivf->mp4 conversion only, everything works for -c:v av1
[20:41:09 CEST] <TD-Linux> in fact I don't get any samples either for ivf->mp4, hence why it doesn't write a trak.
[20:41:37 CEST] <jamrial> oh, that's because you can't remux av1 from ivf without a parser
[20:41:45 CEST] <jamrial> ivf has no way to flag keyframes
[20:42:51 CEST] <TD-Linux> okay. earlier ivf->mp4 was "working" where it marked each sample as a keyframe, which is what I was testing with. it's okay that it doesn't work now though.
[20:43:34 CEST] <durandal_1707> does av1 have standard extradata?
[20:44:16 CEST] <TD-Linux> there's work to get the mp4 and matroska extradata to match
[20:44:50 CEST] <jamrial> ivf -> mp4 never worked unless you use -copyinkf
[20:45:02 CEST] <jamrial> and that creates faulty output, i think
[20:45:11 CEST] <jamrial> an av1 parser is needed
[20:45:28 CEST] <jamrial> i'm not sure how it could have worked for you
[20:45:38 CEST] <durandal_1707> who gonna write little parser?
[20:45:45 CEST] <TD-Linux> huh, I definitely had it working... maybe the incorrect AV01 tag also caused it to have some default codec behavior
[20:46:22 CEST] <TD-Linux> anyway I don't really care because it was still wrong without a parser.
[20:46:55 CEST] <jamrial> did you try my suggestion to force the av01 tag?
[20:47:05 CEST] <atomnuker> I'm getting fed up with these piece of shit incomplete dnn patches
[20:47:13 CEST] <atomnuker> its all just fucking promises
[20:47:29 CEST] <atomnuker> WHYYYYY  DID WE MERGE THEM
[20:47:38 CEST] <TD-Linux> jamrial, yeah it works, I haven't posted a patch yet (it's in addition to the existing one on the list, which fixes MOV)
[20:47:53 CEST] <jamrial> atomnuker: because nobody but the mentor reviewed them for the entire gsoc period
[20:48:07 CEST] <atomnuker> jamrial: even the fucking mentor didn't do his fucking job
[20:48:19 CEST] <atomnuker> michaelni: perdo shouldn't have commit rights!
[20:48:28 CEST] <jamrial> atomnuker: calm down
[20:49:12 CEST] <jamrial> if you want to request a change, do it tomorrow after you chilled down
[20:49:19 CEST] <durandal_1707> they told that they think how to remove tables from repo
[20:49:37 CEST] <jamrial> if you write an email now, you'll do it while pissed and we'll all regret it later
[20:55:16 CEST] <kierank> durandal_1707: why that bit commented
[20:57:18 CEST] <durandal_1707> dunno, its very fragile sync code
[20:57:30 CEST] <cone-931> ffmpeg 03Mina 07master:729ebd3cd001: lavfi/vf_colorconstancy: change option ranges
[20:57:31 CEST] <cone-931> ffmpeg 03Mina 07master:f5559dd8694c: lavfi/vf_colorconstancy: cosmetic update
[20:58:50 CEST] <durandal_1707> need to look what vlc and mplayer do
[21:02:27 CEST] <pasouza> atomnuker: I asked Sergey to prepare a patch removing all the data, it will be ready soon
[21:05:00 CEST] <atomnuker> what, another patch which removes some of it and replaces most of it like the last one?!?
[21:08:32 CEST] <pasouza> the removal depends on the previous patches that were just push
[21:10:05 CEST] <pasouza> a few days ago
[21:10:35 CEST] <pasouza> be sure all the data will be removed
[21:11:54 CEST] <pasouza> regarding the swscale conversion in the filter I can't see how it could be done without doing useless computation in the middle
[22:49:30 CEST] <pasouza> I'm just reading the back log and I'm really done, I tried to be most polite possible and address all you requests, btw the student is addreesing all the requests, he sent 8 patches (12 days ago) and I did not see any word from you in any of them. You said you would personally review the work and you failed to do so. If you think they are pieces of incomplete shit why you did not say so on the review?
[22:50:03 CEST] <pasouza> if you think my review is not good enough at least tell me what to improve, other way I wll continue as I am
[22:51:32 CEST] <pasouza> I'm tired of spending my time and receive these harsh emails and comments
[22:52:24 CEST] <TD-Linux> jamrial, did you see https://github.com/AOMediaCodec/av1-isobmff/issues/71
[23:59:34 CEST] <atomnuker> pasouza: I've been monitoring and hoping that gradually the shit would get resolved
[23:59:56 CEST] <atomnuker> it still hasn't and I said from the start, it would have been far easier to revert
[00:00:00 CEST] --- Wed Aug 15 2018

More information about the Ffmpeg-devel-irc mailing list