[Ffmpeg-devel-irc] ffmpeg.log.20161022

[burek]

[00:39:57 CEST] <Phi> hey folks
[02:25:09 CEST] <{{{}}}> hey there, im on linux mint and i followed this guide to build ffmpeg: https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu
[02:26:28 CEST] <{{{}}}> when i tried to use the play() function in python using pydub, it defaults to ffmpeg and i get this error:     PATH="$HOME/bin:$PATH" PKG_CONFIG_PATH="$HOME/ffmpeg_build/lib/pkgconfig" ./configure \ Invalid Syntax
[02:28:20 CEST] <t4nk860> Hi, I have a small question regarding lossless compression using x264 (or x265,vp9):
[02:29:48 CEST] <t4nk860> I used `ffmpeg -i $input -c:v libx264 -crf 0 $output`
[02:29:52 CEST] <t4nk860> for lossless compression
[02:30:19 CEST] <t4nk860> However, the resulting frames seem to be different (different output for some image processing algo)
[02:30:33 CEST] <t4nk860> Does crf 0 not produce perfectly lossless?
[02:30:55 CEST] <t4nk860> If not, what can I do to obtain lossless performance?
[02:31:08 CEST] <furq> use -qp 0 if you want to be sure it's lossless
[02:31:14 CEST] <furq> -crf 0 is only lossless with 8-bit output
[02:32:00 CEST] <furq> you can also check the frames are identical with the framehash muxer
[02:32:26 CEST] <t4nk860> Thanks for the feedback. I will try with qp 0
[02:36:45 CEST] <DHE> t4nk860: there's still colourspace conversion loss before x264 even gets the frames. consider libx264rgb or yuv444 mode depending on what you're up to exactly.
[02:37:17 CEST] <furq> it should default to yuv444p if the input is rgb
[02:41:19 CEST] <{{{}}}> okay woops i figured out why i got that error. i have a new one now.
[02:41:29 CEST] <{{{}}}> it isnt finding ffplay, do i need to install that package seperately?
[02:42:14 CEST] <t4nk860> DHE: Thanks for pointing this out. Yes, I am manually specifying the format to be yuv444p (even if it might not be necessary)
[02:44:49 CEST] <t4nk860> furq: Will the `-qp 0` flag also work for vp9, x265? ( https://trac.ffmpeg.org/wiki/Encode/VP9 says to explicitly state -lossless 1 flag... but was wondering the original soln will work)
[02:45:15 CEST] <furq> i have no idea
[02:45:28 CEST] <furq> i doubt it
[02:46:43 CEST] <furq> apparently qp 4 is lossless in x265, but i don't know if -qp is actually mapped to anything in libx265
[02:47:00 CEST] <furq> you might need -x265-params qp=4 (or whatever the syntax is)
[02:47:53 CEST] <furq> -x265-params lossless=1
[02:50:39 CEST] <t4nk860> cool.. ll try it out
[14:37:56 CEST] <beauty> hello
[14:38:21 CEST] <beauty> what's the aligned data and none-aliged data??
[14:38:51 CEST] <JEEB> aligned allocation of data so that the ways it's handled can be optimized
[14:40:05 CEST] <beauty> could you give a example?
[14:48:41 CEST] <DHE> SSE-style instructions work in 128 bit typically. but that means if your memory allocation isn't a multiple of 128 bits your program may crash while doing image processing
[14:51:29 CEST] <beauty> thanks
[15:03:02 CEST] <Phi> moop
[15:03:31 CEST] <Phi> can anyone tell me how to access libx264's internal data while encoding?
[15:03:59 CEST] <Phi> AVCodecCtx::priv_data has null x264 encoder and stuff
[15:04:11 CEST] <Phi> also hi
[15:17:51 CEST] <BtbN> "libx264's internal data"?
[15:36:07 CEST] <DHE> x264 only exposes a configuration data structure with things like bitrate, keyframe interval, image size, etc. and even then there's an API to deal with only the text names/values so you can minimize a lot of xx264 structure data access
[16:01:41 CEST] <Phi> yea DHE, but in this case, I'm passing options to libx264 via libav* and they're being ignored
[16:02:26 CEST] <Phi> to get more debugging info I need to access whatever libav* is accessing, because what it's exposing as output options seems to have no bearing on the *actual* options
[16:03:41 CEST] <Phi> vis a vie, I set profile to "baseline" in any option I can, and it just sets it to high
[16:05:49 CEST] <Phi> from what I understand you can encode a H264 stream from one profile to another, but it's not doing that, it insists on reflecting the input profile/level
[16:08:07 CEST] <Phi> I did see notes that you can't downgrade a profile in x264_param_set_profile(), but the input and output should be completely unrelated
[16:08:46 CEST] <Phi> unless I have to decode every frame to an image, then encode the image to the new stream, I'm at a loss
[16:09:11 CEST] <BtbN> There is no such information carried over at all.
[16:09:23 CEST] <BtbN> x264 takes raw images in, and produces h264.
[16:09:32 CEST] <BtbN> All according to the options you pass it.
[16:09:42 CEST] <Phi> I've passed baseline in every option I can find
[16:09:58 CEST] <Phi> codeccontext->profile, av_opt_set(codecctx->priv_data, "profile" ...
[16:10:17 CEST] <BtbN> likely some other option you pass overrides it.
[16:10:24 CEST] <BtbN> baseline has quite a few constraints
[16:10:39 CEST] <Phi> well, atm I'm just passing baseline, all the other options are defaults
[16:11:47 CEST] <BtbN> what chroma format? What resolution?
[16:12:01 CEST] <Phi> YUV420p, and whatever the RTSP is
[16:12:26 CEST] <Phi> I tried two methods - av_copy_context from the RTSP, and av_context_defaults with manually setting width, height, etc
[16:12:49 CEST] <Phi> http://stackoverflow.com/questions/40053873/ffmpeg-rtsp-stream-to-mpeg4-h264-file-using-libx264
[16:13:01 CEST] <BtbN> encoding baseline works fine for me. You are doing something wrong.
[16:13:25 CEST] <Phi> I hope I am, otherwise the libav* guys are doing something wrong :3
[16:14:23 CEST] <Phi> X264Context *x4 = (X264Context *)video_file_stream->codec->priv_data; // is that valid?
[16:14:33 CEST] <Phi> because x4->enc, the encoder, is always null
[16:15:06 CEST] <BtbN> X264Context is a private struct inside of libx264.c
[16:15:16 CEST] <BtbN> you can't access it from the outside anyway, except for setting AVOptions
[16:15:33 CEST] <BtbN> *via setting AVOptions
[16:15:43 CEST] <Phi> I'm doing read-only trying to read variables for debugging
[16:15:46 CEST] <BtbN> priv_data is called private for a reason
[16:16:17 CEST] <BtbN> https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/libx264.c#L677
[16:16:17 CEST] <BtbN> https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/libx264.c#L729
[16:16:39 CEST] <BtbN> https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/libx264.c#L905
[16:16:47 CEST] <BtbN> looks fine, and works fine, for me.
[16:17:04 CEST] <Phi> not at all for me XD
[16:17:13 CEST] <BtbN> you're doing something wrong then.
[16:18:22 CEST] <Phi> well, the code's in the link I posted
[16:18:57 CEST] <Phi> so to clarify, to set profile I merely change avcodecctx->profile, before I encode the first frame?
[16:21:11 CEST] <Phi> is it required I call avcodec_open before the first frame?
[16:22:58 CEST] <BtbN> Of course you have to open the encoder before you can use it.
[16:23:04 CEST] <Phi> weirdly not
[16:23:16 CEST] <Phi> when I do it tends to error
[16:23:25 CEST] <Phi> and I still get an MP4 without one
[16:23:32 CEST] <Phi> so I'm guessing something is screwed up before then
[16:24:03 CEST] <Phi> the examples I've seen don't include calling avcodec_open
[16:24:18 CEST] <Phi> to clarify, I'm doing it inside an AVFormatContext's stream
[16:27:12 CEST] <Phi> so, it's av_guess_format for MP4
[16:27:27 CEST] <Phi> avformat_alloc_context for the format
[16:27:49 CEST] <Phi> set oformat for that context to the av_guess_format
[16:28:24 CEST] <Phi> set filename for format context to oformat's filename
[16:28:38 CEST] <Phi> avcodec_find_encoder for the formatcontext's video codec
[16:28:56 CEST] <Phi> then call avio_open2 on format context's pb
[16:29:04 CEST] <Phi> then call avformat_new_stream
[16:29:19 CEST] <Phi> then I set options
[16:29:27 CEST] <Phi> then avcodec_open2
[16:29:31 CEST] <Phi> then avformat_write_header
[16:29:36 CEST] <Phi> then av_write_frame to infinity
[16:29:49 CEST] <Phi> is that the right order?
[16:44:00 CEST] <rzz> hi, does anyone have experience in switching audio track from 3d mkv?
[16:44:35 CEST] <rzz> I seem to get lot of stuttering and dropped frames
[16:45:01 CEST] <beauty> https://ffmpeg.org/doxygen/trunk/doc_2examples_2decoding_encoding_8c-example.html
[16:45:11 CEST] <beauty>  fwrite(decoded_frame->data[0], 1, data_size, outfile);
[16:45:32 CEST] <beauty> the audio can't be played in example program.
[16:47:00 CEST] <beauty> Is it beacause the decoded data is AV_SAMPLE_FMT_FLTP ,but the way to write data[0] to file is wrong?
[16:49:56 CEST] <Phi> sorry, new to libav*
[17:52:45 CEST] <AlienCat> hey
[17:53:13 CEST] <AlienCat> I am trying to convert an SILK V3 amr file
[17:53:34 CEST] <AlienCat> ffmpeg -i msg.amr -c:v opus glass.wav
[17:53:42 CEST] <AlienCat> but it doesn't work
[17:54:00 CEST] <AlienCat> Invalid frame size (732): Could not seek to 4174.
[17:54:07 CEST] <AlienCat> got this mp3 error
[18:14:49 CEST] <BtbN> AlienCat, you realize opus is an audio codec?
[18:14:59 CEST] <AlienCat> yea
[18:15:06 CEST] <BtbN> But you are telling it to use it as video codec.
[18:15:07 CEST] <AlienCat> I wanna conver audio
[18:15:35 CEST] <BtbN> and are also telling it to create a wav file, which is raw uncompressed pcm
[18:17:31 CEST] <Phi> BtbN, on a separate note I'm getting av_log passed "libx264 : profile Constrained Baseline, level 3.1", but output is still High
[18:19:43 CEST] <AlienCat> so what should I type?
[18:28:13 CEST] <BtbN> depends on what you want. You want a wav file? Encode it with opus?
[18:32:51 CEST] <AlienCat> I wanna force input opus
[18:33:05 CEST] <AlienCat> then convert it to ogg or something
[18:41:26 CEST] <Phi> BtbN, can you demo it for me?
[18:41:40 CEST] <Phi> I set to Baseline, and it's set to Constrained Baseline, and the output is High
[18:43:41 CEST] <BtbN> ffmpeg -f lavfi -i testsrc2 -c libx264 -profile baseline out.mkv
[18:49:26 CEST] <Phi> that works, I'm going to try the RTSP
[19:01:03 CEST] <BtbN> there is nothing left of the input container/codec, once it reaches libx264
[19:01:37 CEST] <JEEB> other than timestamps :P
[19:08:44 CEST] <Phi> one can only hope
[19:08:53 CEST] <Phi> I still have a dopey system
[19:09:08 CEST] <jpsharp> I need a suggestion on how to get ffmpeg to convert a single frame (but occasionally changing) gif into an rtmp stream.  I can get a static stream going, but it never recognizes the new gif if I change it out.  And if I kill ffmpeg and start a new stream with a new gif, the far end clients freak out.
[19:09:18 CEST] <Phi> the RTSP seems to work, although I'm getting a shedload of "Past duration 0.(some number) too large"
[19:09:58 CEST] <jpsharp> Are you streaming a camera Phi?
[19:10:54 CEST] <Phi> a network one, I'm just outputting to a MP4
[19:11:48 CEST] <Phi> I set the profile to Baseline via options passed to avcodec_open2, if I read back x264's internal codec it's Constrained Baseline, and the resulting file is High
[19:11:53 CEST] <Phi> so it makes no sense
[19:14:03 CEST] <Phi> av_log  shows "libx264 : profile Constrained Baseline, level 3.1"
[19:14:09 CEST] <Phi> output file is *still* High
[19:14:19 CEST] <bencoh> Phi: unless you're passing options to x264 *after* setting profile
[19:14:58 CEST] <bencoh> profile/level should be set last
[19:15:19 CEST] <bencoh> (since it's a constraint)
[19:16:26 CEST] <Phi> I'll double-check, but I don't think so
[19:17:29 CEST] <Phi> hm, I do avcodec_get_context_defaults3, then I set the profile
[19:17:38 CEST] <Phi> then I do avcodec_open2
[19:18:14 CEST] <Phi> I'll drop the defualts call and see if it works
[19:23:38 CEST] <Phi> nope, same
[19:23:58 CEST] <Phi> I set it to baseline (not constrained baseline), libx264 log shows constrained baseline, output file is high
[19:25:35 CEST] <BtbN> are you looking at the right output file? Are you sure you are writing anything?
[19:25:37 CEST] <Phi> makes me want to headbutt the wall
[19:25:45 CEST] <BtbN> does it work with ffmpeg cli?
[19:25:47 CEST] <Phi> it's a new filename every time, incrementing based on existing files
[19:25:58 CEST] <bencoh> try setting profile after _open() ?
[19:25:59 CEST] <Phi> commandline works, yeah
[19:26:16 CEST] <BtbN> after avcodec_open2, it won't have any effect
[19:26:22 CEST] <bencoh> (not sure it which order those should be called tbh)
[19:26:27 CEST] <bencoh> ah nvm then
[19:26:33 CEST] <Phi> I'll try it anyway, won't kill me
[19:26:45 CEST] <BtbN> they are only passed to libx264 during init
[19:26:53 CEST] <bencoh> right
[19:30:14 CEST] <Phi> you guys know how to use avformatcontext?
[19:32:17 CEST] <Phi> yeah, no difference
[19:39:22 CEST] <Phi> avio_open2, then avformat_new_stream, right?
[19:45:30 CEST] <Pet0r> hi - I'm trying to encode an input file with the hevc_nvenc encoder, it seems to work with any -preset option just fine, but if I try to actually use "lossless" or "losslesshp", I get "[hevc_nvenc @ 0000023098f10780] No NVENC capable devices found"
[19:45:42 CEST] <Pet0r> any other presets work fine and are certainly HW accelerated
[19:52:27 CEST] <AlienCat> ffmpeg -f -acodec libopus -i msg.amr glass.ogg
[19:52:43 CEST] <AlienCat> still complains aboput mp3
[19:54:47 CEST] <Pet0r> never mind ,eventually found the answer to my own question - lossless is Pascal only and I'm on Maxwell
[20:13:56 CEST] <Phi> and on my end... the profile is still too damn High
[20:25:55 CEST] <Phi> this is such a trippy program
[20:26:16 CEST] <Phi> BtbN, whyyy
[20:29:12 CEST] <furq> have you disgraced any altars recently
[20:31:19 CEST] <Phi> ...maybe
[20:31:46 CEST] <Phi> I've just noticed "writing application" is Lavf56.1.0 in my program, but from ffmpeg it's libx264
[20:31:56 CEST] <Phi> (in metadata in the mp4s)
[20:32:20 CEST] <Phi> so... even though libx264 is writing to av_log... it's not being used?
[20:37:46 CEST] <HorikawaOtane> Have a mild question.
[20:38:00 CEST] <HorikawaOtane> How would one go about selecting the gpu for scale_npp?
[20:40:19 CEST] <BtbN> it uses the one of the context it gets passed. In itself it doesn't have any context creation logic.
[20:42:49 CEST] <HorikawaOtane> i see
[20:42:55 CEST] <HorikawaOtane> so how would one create a context for it?
[20:43:06 CEST] <HorikawaOtane> is there some global variable i need to pass in or something?
[20:43:11 CEST] <HorikawaOtane> like GPU=1 ffmpeg <stuff>
[20:43:54 CEST] <BtbN> Either cuvid, hwupload_cuda or ffmpeg_cuvid.c does it.
[20:44:13 CEST] <BtbN> They all have their own options for device selection
[20:46:12 CEST] <AlienCat> okay I think I found my problem
[20:46:23 CEST] <AlienCat> there are no opus demuxer
[21:00:33 CEST] <Phi> BtbN, I set the codec to libx264, but the output file looks like it's produced by "Lavf56.1.0"
[21:01:07 CEST] <BtbN> so? lavformat is what creates the output file.
[21:02:31 CEST] <Phi> ffmpeg produces one that says libx264
[21:03:04 CEST] <Phi> ...no, not quite
[21:03:12 CEST] <Phi> I have vastly different output data for them
[21:03:25 CEST] <furq> i was going to say, it shouldn't
[21:03:30 CEST] <furq> the writing application is supposed to be the muxer
[21:03:33 CEST] <furq> which is libavformat
[21:03:47 CEST] <furq> if you wrote it with mkvmerge it'd say mkvmerge
[21:04:22 CEST] <Phi> http://pastie.org/private/la1exrsqbws16tf56a7csa is the faulty one made by my app
[21:04:54 CEST] <Phi> http://pastie.org/private/eof0iwihqsynnzuvis7fua is the ffmpeg one
[21:05:16 CEST] <Phi> my one doesn't even say the writing library
[21:06:55 CEST] <furq> it does
[21:07:14 CEST] <furq> it says a different version of lavf for some undoubtedly fun reason
[21:07:38 CEST] <furq> it's missing the encoding settings though
[21:07:54 CEST] <furq> i thought that was something libx264 dealt with internally
[21:07:58 CEST] <HorikawaOtane> So - another question
[21:08:10 CEST] <HorikawaOtane> Is there any particular reason to use nvresize vs scale_npp?
[21:15:33 CEST] <Phi> furq: it says the writing application, not the writing library :(
[21:15:52 CEST] <furq> oh right
[21:16:17 CEST] <BtbN> nvresize?
[21:16:57 CEST] <HorikawaOtane> yeah
[21:17:18 CEST] <BtbN> Is that some external application?
[21:17:19 CEST] <HorikawaOtane> nvresize is an ffmpeg patch by nvidia - https://developer.nvidia.com/ffmpeg
[21:17:24 CEST] <BtbN> Oh, that.
[21:17:38 CEST] <BtbN> Well, you need to patch ffmpeg with an incredibly messy external patch set.
[21:17:50 CEST] <BtbN> Which I doubt still applies.
[21:17:53 CEST] <HorikawaOtane> oh i guess the key with nvresize is that you can also export to multiple resolutions simultaneously
[21:17:57 CEST] <HorikawaOtane> and nah, it doesn't patch cleanly now
[21:18:25 CEST] <HorikawaOtane> but nvresize also doesn't seem to allow you to set your scaling method...
[21:18:38 CEST] <HorikawaOtane> whereas scale_npp accepts interp_algo
[21:19:45 CEST] <Phi> hmm
[21:20:20 CEST] <Phi> I want to test by someone else writing an app
[21:20:26 CEST] <Phi> but that'll take ages
[21:22:17 CEST] <BtbN> nvresize is a pre-compiled cuda kernel which does the scaling with exactly one algorithm.
[21:22:34 CEST] <BtbN> No idea what Algorithm it even is
[21:22:39 CEST] <BtbN> Something simple I'd guess
[21:27:05 CEST] <Phi> sempaiii
[00:00:00 CEST] --- Sun Oct 23 2016