[Ffmpeg-devel-irc] ffmpeg-devel.log.20180108

burek burek021 at gmail.com
Tue Jan 9 03:05:03 EET 2018


[00:03:27 CET] <FishPencil> Is the frame data format always the same by the time it gets to filter_frame()? I think that format is rgb24?
[00:03:52 CET] <nevcairiel> your filter can determine which formats it supports
[00:09:46 CET] <FishPencil> Hm ok, is AVPixFmtDescriptor generally used then to handle the multiple formats?
[00:10:04 CET] <FishPencil> I see most filters don't write specific colorspace functions
[00:16:41 CET] <durandal_170> FishPencil: yes, colorspace conversion is auto done by swscale
[00:17:41 CET] <cone-494> ffmpeg 03Carl Eugen Hoyos 07master:6874945fbfcd: fftools/ffmpeg_opt: Remove a write-only variable.
[00:51:25 CET] <cone-494> ffmpeg 03Mark Thompson 07master:202b59cd3d01: lavfi/opencl: Use filter device if no input device is available
[00:51:26 CET] <cone-494> ffmpeg 03Mark Thompson 07master:dfdc14616112: lavfi: Add filters to run arbitrary OpenCL programs
[00:51:27 CET] <cone-494> ffmpeg 03Mark Thompson 07master:03c08d59fb2b: doc/filters: Document OpenCL program filters
[00:51:48 CET] <wm4> nice
[01:28:07 CET] <wm4> we really should add AVFrame.media_type and AVFrame.time_base fields
[01:29:18 CET] <BtbN> A Patch from Rynair oO
[01:29:37 CET] <BtbN> (Except that there is no patch in the mail)
[01:31:06 CET] <Compn> (as expected from a 3rd rate airline, fails to deliver)
[01:31:14 CET] <Compn> ehe
[01:32:23 CET] <jdarnley> nice
[01:34:54 CET] <wm4> but an impressive disclamer laywer blurb
[01:35:01 CET] <wm4> *disclaimer
[01:35:10 CET] <wm4> no doubt added by their mail system, maybe it even filtered the patch
[01:40:22 CET] <wm4> now he resent from a gmail address
[01:56:41 CET] <RiCON> and still with no patch
[01:56:51 CET] <RiCON> oh no, it's in a separate email
[01:58:33 CET] <Compn> https://en.wikipedia.org/wiki/FFmpeg still lists ffserver :(
[02:12:19 CET] <jamrial> it will be edited eventually. or now if you do it yourself :p
[02:18:38 CET] <wm4> does anyone mind if I push that network init change patch?
[03:22:58 CET] <jamrial> wm4: it's three days old, and two of those days were a weekend. give it a couple more
[04:52:51 CET] <cone-869> ffmpeg 03James Almer 07master:9d68c0191cdd: fate/filter-video: fix 12 bit framerate filter tests on big endian targets
[07:38:47 CET] <FishPencil> Something seems odd about the nlmeans filter. I'm comparing it to the reference filter from A. Buades himself and it doesn't even look like it's operating the same way.
[08:45:41 CET] <cone-515> ffmpeg 03Paul B Mahol 07master:205046420d5a: avfilter/af_aiir: fix typo which may cause overread
[09:27:24 CET] <kierank> BtbN: ryanair?
[09:28:03 CET] <kierank> oh yeah
[09:28:04 CET] <kierank> lol
[11:25:44 CET] <thardin> could I get someone to take a look at my codec2 patchset?
[11:43:31 CET] <cone-515> ffmpeg 03Paul B Mahol 07master:3f234a0b22bc: avfilter/af_aiir: do not forget to free gains too
[11:48:43 CET] <durandal_170> when are documentation pages updated on ffmpeg.org?
[11:51:59 CET] <BtbN> I think it's a cronjob that pulls latest master every once in a while
[12:20:59 CET] <lozarythmic> Hello! I'm trying to find out some information on how conversion is done. One of my co-workers has been extracting color frames to grayscale by extracting the Y pixel values from a video and exporting them to a frame. I think that '-vf format=gray' does pretty much the same thing but I'd like to have a look at the code that does it.
[12:21:39 CET] <lozarythmic> Unfortunately while I am a bit of a coder I'm not really into C and was hoping one of you gurus could point me at the right file? I've had a look through the some of the code on GitHub but I'm really just poking around in the dark
[12:52:49 CET] <Compn> lozarythmic : 
[12:53:00 CET] <Compn> its so early :D
[12:53:45 CET] <Compn> for that specific vf the file is in libavfilter
[12:54:09 CET] <Compn> lozarythmic : https://github.com/FFmpeg/FFmpeg/blob/master/libavfilter/af_aformat.c
[12:54:36 CET] <Compn> but that file wont help you much as it calls other code
[12:54:43 CET] <Compn> err wrong file
[12:55:05 CET] <Compn> https://github.com/FFmpeg/FFmpeg/blob/master/libavfilter/vf_format.c
[12:58:51 CET] <SortaCore> I'm now manually going through all the ffmpeg.c code to work out the proper way to time streams
[12:59:50 CET] <SortaCore> the variable naming in this is making me slightly on the murderous side
[13:00:00 CET] <lozarythmic> Thanks Compn - it's GMT here so almost lunchtime :)
[13:00:33 CET] <SortaCore> "stime", "stime2", "st", "ist->st", "start_time"
[13:01:48 CET] <Compn> SortaCore : you may not like what you see... but if you make a patch to unify it, that might also be nice
[13:02:37 CET] <Compn> lozarythmic : depending what your coworker is doing, it could be the same, or different. there are a few ways to do greyscale conversion iirc
[13:03:47 CET] <SortaCore> you're asking someone who can't understand what he reads to write something better based off it?
[13:04:19 CET] <SortaCore> I'll just do __debugbreak(); and let the coder manually change the variables
[13:05:47 CET] <SortaCore> this will result in cleaner, more concise code
[13:06:15 CET] <SortaCore> might execute a bit slowly but that's pebkac right
[13:07:33 CET] <lozarythmic> Compn: here's an extract from coworkers script: -pix_fmt yuv422p $outdir/$framefile.%9d.Y
[13:09:08 CET] <Compn> yuv422p is a standard pix fmt with color :P
[13:09:21 CET] <lozarythmic> Unfortunately he's no longer with the company or i'd ask him!
[13:09:48 CET] <Compn> haha
[13:10:13 CET] <Compn> lozarythmic : well then try comparing his old script to ffmpeg gray output and see if its the same
[13:10:24 CET] <Compn> lozarythmic : but since this is a user question ... you might want to try asking in #ffmpeg
[13:10:49 CET] <Compn> this channel is for development of ffmpeg code not 3rd party scripts and such
[13:10:51 CET] <lozarythmic> A couple of lines down in the script he adds a header to that .Y file and turns it into a .pgm: echo "P5\n$width $height\n255" >$yfname.pgm
[13:11:34 CET] <lozarythmic> I did wonder if that was the case. It was mainly because i was looking through the code in github that i ende dup here. But I shall ask there. Thanks for your time
[13:11:43 CET] <Compn> yep good luck
[13:13:52 CET] <SortaCore> generally few people who do C++ are on #ffmpeg though...
[13:14:13 CET] <SortaCore> also, InputStream struct alone has about 26 uncommented variables
[13:17:05 CET] <Compn> SortaCore : but you understand we are entitled to have our own channel for development yes? :P
[13:19:43 CET] <SortaCore> yea, no problem
[13:20:05 CET] <SortaCore> but you get diverted back to ffmpeg and then get little help :p
[13:20:49 CET] <SortaCore> several times, I've asked multiple times on ffmpeg and got nothing (at various points in the day), and on devel I get responses because people here are all about the C++
[13:20:53 CET] <cone-515> ffmpeg 03Carl Eugen Hoyos 07master:e8f7171a342e: lavf/oggparseogm: Remove an unneeded assert.
[13:21:08 CET] <Compn> its almost like ffmpeg developers arent paid to help random people with random c++ questions
[13:21:16 CET] <Compn> ehe
[13:22:03 CET] <SortaCore> they're not paid to help anyone anytime :p
[13:22:49 CET] <SortaCore> speaking of which, that notice about the net neutrality vote is still on the ffmpeg.org site
[13:22:59 CET] <Compn> i didnt put it there :D
[13:23:13 CET] <SortaCore> I wouldn't know, what do you work on?
[13:23:20 CET] <Compn> maybe i should fix it
[13:24:52 CET] <SortaCore> I just have this massive user list here and no idea who works on what
[13:25:04 CET] <SortaCore> is there a document about it
[13:25:26 CET] <Compn> i'll answer this question for $20
[13:25:47 CET] <nevcairiel> i'll answer it for free: "no"
[13:26:15 CET] <nevcairiel> everyone is free to work on anything they want, and they dont have to register with anyone =p
[13:27:45 CET] <Compn> i mean there is a file detailing who works on what though :P
[13:28:19 CET] <SortaCore> yea, I see people going for approval to other users
[13:28:23 CET] <SortaCore> for patches and the like
[13:28:31 CET] <Compn> like i said, $20
[13:29:09 CET] <Compn> or 20 bitcoins
[13:29:15 CET] <nevcairiel> that "file" isnt particularly helpful
[13:29:31 CET] <nevcairiel> mostly its "what did people work in the past, maybe even 10 years ago"
[13:32:37 CET] <SortaCore> I'd send you my bitcoins but that might start the bubble
[13:33:26 CET] <SortaCore> instead I'll hang around making snarky remarks
[13:33:36 CET] <SortaCore> snarky
[13:37:00 CET] <j-b> how much is BTC today? $10k or $25k?
[13:37:31 CET] <Compn> btc is a giant ponzi scheme... get in and out , dont get left holding the empty bag...
[13:38:00 CET] <Compn> j-b : was $15k last night
[13:38:37 CET] <SortaCore> heard Ethereum is the new thing
[13:38:46 CET] <Compn> all coins are ponzi schemes
[13:39:20 CET] <SortaCore> apparently some of the btc creators left it as btc was being treated like a material, not like currency
[13:39:41 CET] <Compn> or the new word for ponzi scheme "money stores / store of value" 
[13:40:33 CET] <nevcairiel> a ponzi scheme is a rather specific way to exploit people, a coin by itself can't be that - although many people seem to have setup such schemes promising huge gains on top of cryptocoin hype
[13:41:17 CET] <Compn> nevcairiel : so you would say btc is a ponzi or not ?
[13:41:27 CET] <Compn> complete with bitcoin cash now... lol
[13:42:43 CET] <nevcairiel> its not, its like  any highly volatile stock or even a traditional currency, of course you can be left holding the bag if you dont pay attention, but that can happen with any stock trading or such as well
[13:43:42 CET] <nevcairiel> what one has to be wary about is companies offering you fixed gains over short periods of time by claiming to play crypto currencies for you, those can easily be fraudulent
[13:44:15 CET] <Compn> its nothing like a traditional currency
[13:44:28 CET] Action: Compn runs
[13:44:43 CET] <atomnuker> Compn: you're biased, you do stock trading
[13:44:52 CET] <Compn> i dont like stock trading
[13:44:59 CET] <nevcairiel> of course not, but speculating it  is sort of the same thing
[13:45:05 CET] <nevcairiel> except cryptocoins are more volatile
[13:45:15 CET] <SortaCore> I'm seeing some stuff about cryptocoins in the ponzi scheme wiki article
[13:45:19 CET] <Compn> the speculation market is the same yes
[13:45:52 CET] <SortaCore> speaking of speculation, processors, amirite
[13:48:16 CET] <Compn> bitcoin is going to be banned because its not taxed...
[13:48:52 CET] <thardin> profits from it are taxed, at least in sweden
[13:50:47 CET] <Compn> and not regulated like any stock, currency, or bank 
[13:51:01 CET] <nevcairiel> its a bit tricky here, if you leave your money in bitcoin for over a year, i believe its tax free, but if you do shorter speculation, its taxed
[13:51:57 CET] <Compn>  wow your countries have some odd laws :D
[13:52:04 CET] <Compn> probably some corruption going on there heh
[13:53:06 CET] <nevcairiel> its a generic law for selling and buying generic "property"
[13:53:15 CET] <durandal_170> Compn: why are you so clueless about ffmpeg?
[13:53:39 CET] <nevcairiel> if you own it for a year, you can sell it again without extra taxes - if you buy and sell in shorter time frames, you get to pay extra tax
[13:53:52 CET] <nevcairiel> its because they dont recognize it as money yet
[13:55:04 CET] <thardin> I don't think a tax authority would want to consider it money, since profits on money speculation is tax free
[13:55:07 CET] <thardin> at least here
[14:00:29 CET] <Compn> durandal_170 : come to vdd
[14:01:50 CET] <durandal_170> atomnuker: i thinking about fast audio denoiser, which is basically median of size 3
[14:02:58 CET] <durandal_170> Compn: vdd? vast deserted devs?
[14:04:03 CET] <atomnuker> durandal_170: won't be as good as the neural network denoise, just wait for me to finish atrac9
[14:05:59 CET] <durandal_170> atomnuker: but it will be super fast
[14:14:41 CET] <kierank> atomnuker: michaelni: do you know what's going on here: https://git.videolan.org/?p=ffmpeg.git;a=commitdiff;h=37941878f193a2316c514bd5ba55bfe9d2dfdfcf
[14:16:17 CET] <funman> also see https://trac.ffmpeg.org/ticket/5718
[14:31:13 CET] <atomnuker> durandal_170: the neural network denoiser is super fast already
[14:32:17 CET] <atomnuker> kierank: what do you mean?
[14:32:31 CET] <kierank> why is libopus channel_layout broken
[14:33:33 CET] <durandal_170> atomnuker: median is used in declicker, does neurons can fix such signals?
[14:34:05 CET] <atomnuker> durandal_170: yep, anything that affects *voice* only though
[14:34:16 CET] <atomnuker> if you call it a declick filter I'll be fine with it
[14:34:30 CET] <atomnuker> we don't have one afaik and people need to filter vynyls
[14:35:24 CET] <atomnuker> kierank: not sure, if you can reproduce try setting -mapping_family 255
[14:36:00 CET] <atomnuker> (opus has at best muddy definition of layouts and what needs to get paired with what, even atrac9 specified that nicely)
[14:36:39 CET] <atomnuker> and I don't know how mapping_family 255 works to enable arbitrary layouts or how it gets them right (if it does get them right)
[14:37:28 CET] <cone-515> ffmpeg 03Carl Eugen Hoyos 07master:ed03fad2cc52: lavf/oggparseogm: Remove an unneeded include.
[16:07:07 CET] <jkqxz> ...
[16:07:20 CET] <jkqxz> Why do people even try to use qsvdec?  I should write some documentation which warns against it more strongly.
[16:08:06 CET] <nevcairiel> why not just delete it then
[16:12:16 CET] <jkqxz> I was wanting to add external delay before doing that, so that people can't complain that dxva2/vaapi appears slower to naive measurement.
[16:12:51 CET] <nevcairiel> its unfortunate they broke d3d11
[16:12:56 CET] <nevcairiel> i would probably use that
[16:13:17 CET] <nevcairiel> do we have mapping from d3d11 to dxva2? :D
[16:13:52 CET] <jkqxz> Not currently.  Is that possible?
[16:14:16 CET] <nevcairiel> the other way around at least, if you use d3d9ex
[16:14:20 CET] <nevcairiel> not sure that way
[16:28:28 CET] <RiCON> nevcairiel: rossy added dxva2 support to mpv's d3d11 backend recently https://github.com/mpv-player/mpv/commit/88c29b13011175b77c8255d82d5fc1c1610345d5
[16:29:00 CET] <nevcairiel> well like I said, you can share dxva2 surfaces into d3d11
[16:29:05 CET] <nevcairiel> but the other way around, i'm not sure
[16:37:27 CET] <jkqxz> That appears to be doing a copy from the decoded-to d3d9 surface to another d3d9 surface which was created in some shared way from a d3d11 surface?
[16:40:01 CET] <nevcairiel> yeah but on-gpu copy is at least fast
[16:41:52 CET] <nevcairiel> presumably you could just do the copy in the other direction using the same sharing method
[16:45:02 CET] <jkqxz> That does suggest it might work, though, since the real mapping is in the direction you want from d3d11 to d3d9 (create D3D11 texture -> get DXGI resource -> get shared handle -> create D3D9 texture from shared handle -> extract D3D9 surface from texture).
[16:46:30 CET] <jkqxz> Not entirely clear how that would go into hwcontext.  Maybe do the above operation on all the textures during context-derivation, then map is just the synchronisation step and return the other one you already made?
[16:49:26 CET] <wm4> how about intel just fixes d3d11->qsv
[16:53:45 CET] <SortaCore> along with their speculative execution *smashes drum kit*
[16:58:19 CET] <wm4> I mean, the qsv thing is just a driver issue
[19:33:46 CET] <durandal_170> atomnuker: by clicks you mean vinyl, but that is actually recorded sound so nothing to fix
[19:44:17 CET] <cone-515> ffmpeg 03Michael Niedermayer 07master:b3192c64b5bd: avcodec/jpeg2000dsp: Fix integer overflows in ict_int()
[21:01:09 CET] <atomnuker> durandal_170: most of the vinyl rippers from the scene do quite a lot actually
[21:01:29 CET] <atomnuker> declicking on track transitions/quiet parts is extremely common
[22:06:00 CET] <atomnuker> BBB: how does vf_colorspace handle out of gamut colors when e.g. converting XYZ to RGB?
[22:06:13 CET] <BBB> it probably clips it
[22:32:14 CET] <lomancer1> I am trying to understand how libavcodec initializes a AVCodecContext->hwaccel_context for VAApi can anyone point me to where that is initialized in ffmpeg source?
[22:33:27 CET] <wm4> it doesn't do that
[22:33:39 CET] <lomancer1> ffmpeg?
[22:33:43 CET] <lomancer1> what does then?
[22:33:51 CET] <wm4> you can make it automatically manage the context, but it won't set that field
[22:33:55 CET] <wm4> the API user
[22:34:00 CET] <wm4> but that's old API
[22:34:09 CET] <nevcairiel> hwaccel_context is the old way to use vaapi, new code should use hwframes_context
[22:34:15 CET] <nevcairiel> or whatever the member is called
[22:34:23 CET] <nevcairiel> hwframesctx?
[22:35:05 CET] <wm4> hw_frames_ctx or hw_device_ctx (whatever you prefer)
[22:35:37 CET] <wm4> latter for simple uses, former for more advanced ones
[22:36:12 CET] <lomancer1> I tried that and using the hw_decode.c example I am getting an error 
[22:36:15 CET] <lomancer1>    Hardware acceleration context (hwaccel_context) does not exist
[22:38:29 CET] <wm4> that looks like it should work
[22:38:41 CET] <wm4> but the error message is not expected
[22:39:26 CET] <wm4> is that with ffmpeg git master?
[22:39:44 CET] <lomancer1> no the most recent release
[22:39:51 CET] <lomancer1> i can pull and build master
[22:40:02 CET] <wm4> I have no idea what it's like in the release
[22:40:16 CET] <lomancer1> ok will pull and build master then and get back to you
[22:40:26 CET] <nevcairiel> 3.4 should not be much different
[22:40:28 CET] <nevcairiel> its still quite new
[22:40:32 CET] <lomancer1> is it possible that that is an old ffmpeg error message?
[22:41:03 CET] <lomancer1> i still have older versions of ffmpeg on the machine and its possible that is a LD_LIBRARY_RUN time versus link time thing
[22:41:21 CET] <jamrial> 3.4 is missing these couple changes wm4 needed for mpv
[22:41:31 CET] <lomancer1> mpv?
[22:42:14 CET] <wm4> I think the release still has support for the old vaapi API (but also the new one), and the error message you showed is in the old removed code
[22:42:22 CET] <cone-515> ffmpeg 03Michael Niedermayer 07master:1bcd7fefcb3c: avcodec/opus_parser: Check payload_len in parse_opus_ts_header()
[22:42:23 CET] <cone-515> ffmpeg 03Michael Niedermayer 07master:eaa931758955: avcodec/diracdec: Fix integer overflow with quant
[22:42:24 CET] <cone-515> ffmpeg 03Michael Niedermayer 07master:0e62a2373475: avcodec/dirac_dwt: Fix overflows in COMPOSE_HAARiH0/COMPOSE_HAARiL0
[22:42:25 CET] <cone-515> ffmpeg 03Michael Niedermayer 07master:d6945aeee419: avcodec/h264addpx_template: Fixes integer overflows
[23:27:48 CET] <durandal_1707> i did serial biquad cascading \o/
[00:00:00 CET] --- Tue Jan  9 2018


More information about the Ffmpeg-devel-irc mailing list