[Ffmpeg-devel-irc] ffmpeg-devel.log.20160912

[burek]

[00:20:34 CEST] <cone-727> ffmpeg 03Jan Sebechlebsky 07master:b7760dd8d437: avformat/tee: Use BSF list API
[01:18:15 CEST] <cone-727> ffmpeg 03Jens Ziller 07master:bc7066fd5dcf: libavcodec/mmaldec.c: set AVFrame interlaced_frame and top_field_first, AVCodecContext framerate
[01:35:29 CEST] <AnonBaiter> umm
[01:35:41 CEST] <AnonBaiter> did anyone take a look at my 6-channel at3/msf issue?
[01:35:56 CEST] <AnonBaiter> https://www.sendspace.com/filegroup/zKTy7nGQXwdYxVCHvP6Fnw
[01:37:16 CEST] <Chloe> AnonBaiter: what was the issue?
[01:37:40 CEST] <AnonBaiter> see the link
[01:37:51 CEST] <Chloe> yes, there's two iels
[01:37:55 CEST] <Chloe> files*
[01:37:59 CEST] <AnonBaiter> the 6-channel at3 file can't be decoded
[01:38:06 CEST] <AnonBaiter> neither the msa/msf file
[01:38:12 CEST] <Chloe> Is there a ticket for them?
[01:38:16 CEST] <AnonBaiter> nope
[01:38:34 CEST] <Chloe> You should probably make a ticket
[01:38:45 CEST] <AnonBaiter> but I don't have an account?
[01:38:52 CEST] <Chloe> just make one
[01:39:30 CEST] <AnonBaiter> okay
[01:39:41 CEST] <AnonBaiter> oh and how can I san a codec of an input file?
[01:42:05 CEST] <RiCON> http://trac.ffmpeg.org/ticket/5334 use this as an example
[01:49:13 CEST] <AnonBaiter> got it
[01:50:54 CEST] <RiCON> those should probably be two separate issues
[01:51:45 CEST] <Chloe> Yes, edit the ticket to only be for at3, and then make another issue for msf
[01:52:34 CEST] <Chloe> also attach the samples to the ticket
[01:53:09 CEST] <AnonBaiter> but how should I attach the sample?
[01:53:30 CEST] <AnonBaiter> wait
[01:54:10 CEST] <Chloe> scroll down a little, there's 'attachments'
[01:54:40 CEST] <AnonBaiter> got it
[09:57:28 CEST] <cone-926> ffmpeg 03Matthieu Bouron 07master:025f75c3ad1a: lavc/ffjni: do not error out if the last non-mandatory field/method cannot be found
[12:02:38 CEST] <durandal_17> who were on VDD FFmpeg tehnical discussion?
[12:05:01 CEST] <durandal_17> michaelni: libavcodec/h264_refs.c:372:13: warning: variable 'i' is used uninitialized whenever switch default is taken [-Wsometimes-uninitialized]
[12:11:45 CEST] <jkqxz> It can't be outside { 0, 1, 2 } unless you've corruped the state somehow.  Therfore, change to av_assert0?
[12:37:50 CEST] <cone-926> ffmpeg 03Michael Niedermayer 07master:5bbffe3412c1: avcodec/h264_refs: Change default case to av_assert0() to suppress warning
[13:30:57 CEST] <ubitux> http://b.pkh.me/bears_noisy.jpg http://b.pkh.me/bears-denoised.png yey
[13:33:05 CEST] <durandal_17> ubitux: that's not real noise
[13:33:59 CEST] <ubitux> do you have picture with real noise?
[13:34:16 CEST] <ubitux> i guess i could get one from a raw photo
[14:39:57 CEST] <cone-926> ffmpeg 03Paul B Mahol 07master:dc669d5fbed2: avfilter/vf_vaguedenoiser: change thresholding to function pointers
[14:39:58 CEST] <cone-926> ffmpeg 03Paul B Mahol 07master:b791d0e4e0d1: avfilter/vf_vaguedenoiser: calculate dimensions at init
[14:39:59 CEST] <cone-926> ffmpeg 03Paul B Mahol 07master:afeffd891f58: avfilter: add sobel and prewitt filter
[14:40:00 CEST] <cone-926> ffmpeg 03Paul B Mahol 07master:a50ca6b3570f: avfilter/vf_convolution: add slice threading support
[14:46:50 CEST] <durandal_17> atomnuker: you were on VDD FFmpeg tehnical discussion?
[15:36:08 CEST] <j-b> m
[15:43:06 CEST] <durandal_17> hey! ffmbc still have libmpcodecs
[17:14:58 CEST] <durandal_17> does anybody know mp4 analyzer which displays file format structure and it works on Linux?
[17:19:53 CEST] <mateo`_> durandal_17: lsmash
[17:22:15 CEST] <mateo`_> it provides a tool called boxdumper
[17:30:15 CEST] <durandal117> why ffmpeg puts 1904 as creation year in mp4?
[17:38:19 CEST] <nevcairiel> 1904 is the epoch used in mp4 dates, so its likely a zero timestamp
[17:39:12 CEST] <JEEB> someone thought it was a good "anonymous" timestamp at some point
[18:57:02 CEST] <atomnuker> durandal_1707: yeah, ask Compn for the stuff we wrote on the whiteboard when he's online
[18:57:25 CEST] <atomnuker> (I can't remember exactly what we wrote)
[18:59:24 CEST] <Chloe> atomnuker: https://drive.google.com/drive/folders/0B53tgCAS2uUnVUpZN0U1WDE0MzQ?usp=sharing
[19:01:48 CEST] <JEEB> interesting
[19:02:43 CEST] <Chloe> I think updated examples should definitely be a priority
[19:04:38 CEST] <JEEB> true
[19:18:46 CEST] <durandal117> atomnuker: but who wrote/said that we need 'waveforms' and/or 'rgb levels' ?
[19:24:43 CEST] <atomnuker> Compn wrote on the board, maybe he remembers
[19:25:12 CEST] <atomnuker> Yayoi maybe since I think she deals with that
[19:33:05 CEST] <durandal117> atomnuker: and she was at VDD 2016?
[19:33:55 CEST] <JEEB> yes
[19:35:06 CEST] <durandal117> JEEB: you were too?
[19:35:13 CEST] <JEEB> yup
[20:12:27 CEST] <FernetMenta> Hi, can anybody help me with swr_set_compensation?
[20:24:45 CEST] <durandal117> FernetMenta: what's problem?
[20:25:31 CEST] <FernetMenta> is it supposed to be called on every cycle or just after parameters changed
[20:26:17 CEST] <FernetMenta> we (Kodi) observe corrupted audio if we call swr_set_compensation on every cycle with 0
[20:26:43 CEST] <FernetMenta> I would have thought that calling it with zero should just disable it, but this seems not to be the case
[20:28:37 CEST] <FernetMenta> corrupted audio in this context means not bit perfect
[20:29:35 CEST] <durandal117> michaelni: ^
[20:43:32 CEST] <michaelni> FernetMenta, you dont need to call it unless you want some non zero compensation and in that case a single call is enough. Not sure though why calling it with 0 causes a problem
[20:55:38 CEST] <FernetMenta> hmm, then there must be some problem with it. I called it only if parameters changed. audio of a 25 fps movie played back at 24hz speed did not sound good: "geleiher"
[20:56:04 CEST] <FernetMenta> after having it reverted to be called on every resample cycle the problem disappeared
[20:56:48 CEST] <FernetMenta> https://github.com/xbmc/xbmc/commit/8ae935452744c6b655bafe73a6fc1190d0148300
[20:57:07 CEST] <FernetMenta> ^^any isea why this fixed the issue?
[20:57:16 CEST] <FernetMenta> isea = idea
[21:07:14 CEST] <TD-Linux> >VMAF filter
[21:07:26 CEST] <TD-Linux> ... good luck :)
[21:20:13 CEST] <michaelni> FernetMenta, one call will compensate for the specified distance. If theres a persistent error it would need to be called repeatly. 
[21:22:08 CEST] <michaelni> is there some wa by which the problem can be reproduced with ffmpeg ?
[21:36:39 CEST] <FernetMenta> thanks, this answers my primary question. in our case we need to call it repeatly
[21:37:05 CEST] <FernetMenta> I'll think about a test case for replicating the 2nd issue with ffmpeg
[21:37:10 CEST] <wm4> in my code I explicitly compensate for that distance thing so it doesn't do anything
[21:38:19 CEST] <FernetMenta> it is kind of weird, they sent dts audio wrapped by flac, this gets sent as pcm and is only detected by the AVR as dts if untouched
[21:40:01 CEST] <FernetMenta> looks like as soon as a resampler is inited by setting SWR_FLAG_RESAMPLE, audio is slightly altered
[21:53:32 CEST] <TD-Linux> (btw if anyone actually wants to implement different metrics, ping me as I already have implementations and I don't want ffmpeg to have incompatible ones)
[21:54:24 CEST] <atomnuker> psnr-hvs?
[21:56:53 CEST] <durandal_1707> TD-Linux: what you have?
[22:47:50 CEST] <ubitux> so i'm doing patches differences using an ssd integral ((a-b)²); given a patch of size PxP, the maximum difference is P².0xff² with 8-bit material; but now i'm doing a LUT to map these differences to some precalculated values
[22:48:04 CEST] <ubitux> obviously, i need to scale down the difference before doing the mapping
[22:48:24 CEST] <ubitux> (otherwise for a 7x7 patch you get around 3M different possible values)
[22:48:51 CEST] <ubitux> but unfortunately, the meaningful values are in the first entries
[22:49:34 CEST] <ubitux> like, if i'm doing a LUT of size 1024 i'll get something like a dozen values at the beginning and the rest will be zeros
[22:49:59 CEST] <ubitux> (it's mapping a exp(-x))
[22:50:31 CEST] <ubitux> so my question is: with common materiel, what kind of maximum difference i should expect?
[23:13:23 CEST] <ubitux> mmh or maybe i should ask myself how to compute the maximum meaningful difference
[23:15:36 CEST] <durandal117> ubitux: use absolute values instead
[23:16:07 CEST] <ubitux> so i actually need to find a discrete i for exp(-i * C) == 0, or maybe < 0.001 or something
[23:16:12 CEST] <ubitux> durandal117: won't be enough
[23:16:20 CEST] <cone-130> ffmpeg 03Rodger Combs 07master:3b32e1313c6d: ass_split: fix handling of streams with no [Events] or Format: line
[23:16:20 CEST] <cone-130> ffmpeg 03Rodger Combs 07master:dca03ec5f4d5: ass_split: reindent
[23:27:36 CEST] <durandal117> somebody broke mpegts demuxer, I get packets with no timestamps
[23:42:07 CEST] <ubitux> -fflags +genpts? :p
[23:42:19 CEST] <ubitux> at least avi behaves the same
[23:44:28 CEST] <durandal117> doesn't help
[23:46:30 CEST] <ubitux> as input option?
[23:47:23 CEST] <durandal117> yes
[00:00:00 CEST] --- Tue Sep 13 2016