[Ffmpeg-devel-irc] ffmpeg.log.20170304

[burek]

[04:00:00 CET] <slalom> anybody know what the difference is between the 'baseline' profile and 'constrained baseline'?  I'm trying to match some output example data we got as well as possible.. i'm building with -profile:v baseline but instead of the video stream coming out as Video: h264 (Baseline) it outputs as Video: h264 (Constrained Baseline)
[04:03:52 CET] <relaxed> slalom: https://en.wikipedia.org/wiki/H.264/MPEG-4_AVC#Profiles
[04:04:38 CET] <furq> slalom: usually when someone says baseline they mean constrained baseline
[04:04:53 CET] <furq> confusingly there is a separate baseline profile but i don't think anyone really uses it
[04:15:58 CET] <slalom> furq: i see that it's a separate one in that link, but is there any way for ffmpeg to build so that the result actually outputs as h264 (Baseline)?  i mean, i chose baseline
[04:16:49 CET] <slalom> we deploy a 3g-bandwidth version of the media for android, and the example output for that that im trying to match is just Baseline
[04:17:27 CET] <furq> i don't think x264 can do that
[04:17:41 CET] <furq> x264 --profile baseline is constrained baseline
[04:18:02 CET] <furq> you could maybe try a different h264 encoder but i wouldn't know where to start really
[04:18:38 CET] <slalom> damn
[04:18:52 CET] <slalom> ive spent weeks on this and may have to bail on it
[04:19:28 CET] <slalom> i'm basically replacing a pipeline of a producer here where he had to run 15 different encoding presets in adobe media encoder, then copy stuff to a mac and package HLS
[04:19:31 CET] <slalom> i said i could do it all in ffmpeg
[04:19:52 CET] <slalom> everything is working other than the android presets, which are profile baseline level 3.0
[04:20:30 CET] <slalom> still not entirely sure why they're not working yet though, im just investigating differences in the formats at the moment
[04:22:24 CET] <furq> only really old/cheap android devices need baseline
[04:22:34 CET] <slalom> yeah...
[04:22:43 CET] <slalom> i guess that's the failsafe target i dunno
[04:23:29 CET] <furq> also bear in mind that -level in ffmpeg usually doesn't do much
[04:23:44 CET] <slalom> i just used it because i'm copying the adobe media encoder preset values
[04:23:47 CET] <furq> e.g. you can't encode a 720p video with -level 3.0 and get a level 3.0 video
[04:25:58 CET] <furq> annoyingly it will still write the provided level into the sps
[04:26:09 CET] <slalom> weird
[04:26:11 CET] <furq> i guess you could encode it without -level and see what level ffprobe says it is
[04:26:31 CET] <slalom> i need to find out more about how it 'didnt work', he just said it said 'network error' but i dont know if that's a browser error or an error inside our app trying to stream the file
[04:54:04 CET] <c3r1c3-Win> I'm getting the error Error code -22. Error in av_frame_get_buffer in the ffmpeg_acompressor_filter_audio function(failure to allocate frame).
[04:54:28 CET] <c3r1c3-Win> But I can't seem to find any documentation on errors codes.
[06:22:43 CET] <gp5st> `-profile:v baseline -level 3.0 -s 720x486 -b:v 512k -minrate:v 256k -maxrate:v 1.25m -preset veryslow -start_number 0 -hls_time 4 -hls_segment_filename 'if1zwfoxJSs/stream%05d.ts' -hls_list_size 0 -f hls if1zwfoxJSs/stream.m3u8` i'm confused bow -b:v -minrate:v and -maxrate:v interact here. is -b:v the target average? or just what it tries to hit each time, but based on some criteria can change?
[06:30:20 CET] <furq> minrate doesn't actually do anything with x264 afaik
[06:30:32 CET] <furq> although even if it did, it's not actually useful
[06:30:55 CET] <furq> also i'm not sure what maxrate does without bufsize set
[06:31:31 CET] <gp5st> yet more evidence that I don't know anything
[06:32:03 CET] <furq> in short, the average bitrate can't exceed maxrate in every bufsize bytes
[06:32:12 CET] <furq> b:v is just the target bitrate
[06:36:01 CET] <gp5st> are there any other stupid things I'm doing there?
[06:38:25 CET] <furq> there's not really any point having maxrate so much higher than average bitrate
[06:39:23 CET] <furq> if you're prepared to have the bitrate spike that high then you might as well just have the average bitrate be close to that
[06:40:56 CET] <gp5st> hmm
[06:42:01 CET] <gp5st> is it worth even specifying a max then? and just let ffmpeg do it's thing?
[06:42:18 CET] <furq> you probably want to set a maxrate for live streaming
[06:43:25 CET] <furq> if anything you'd want to drop the average bitrate and use crf mode with maxrate/bufsize
[06:44:34 CET] <gp5st> thanks :)
[06:47:19 CET] <gp5st> something else I noticed, I have -hls_time 4 but a lot of my segments are 8s long
[06:52:57 CET] <furq> you probably want to set -g and -x264-params keyint-min
[06:54:02 CET] <furq> just -g is probably ok
[06:59:35 CET] <gp5st> thanks. I saw that but thought -hsl_time would force that/figure it out
[07:08:47 CET] <mritunjay> help,me to get started. i do code in c,tell me from where to move on.
[07:14:47 CET] <c3r1c3-Win> mritunjay: What are you trying to do?
[07:23:36 CET] <mritunjay> sir,i look for "FITS image decoder and encoder,VMAF video filter".
[07:28:53 CET] <c3r1c3-Win> mritunjay: The Google Summer of Code? https://trac.ffmpeg.org/wiki/SponsoringPrograms/GSoC/2017
[07:35:06 CET] <mritunjay> c3r1c3-Win: i have to known how  ,i can start contributing.
[07:35:50 CET] <c3r1c3-Win> mritunjay: The info is on the page that I linked.
[07:35:59 CET] <mritunjay> ok
[07:43:54 CET] <mritunjay> how can i join the mailing lists
[07:44:42 CET] <c3r1c3-Win> mritunjay: The info is on the page that I linked... or rather the page has a link on it to more info.
[08:48:17 CET] <Commander_Cody> An ffmpeg command crashed.
[08:48:23 CET] <Commander_Cody> How could I resume it?
[08:50:31 CET] <Commander_Cody> ffmpeg -i input -ss 1:00:00 -to 1:30:00 -vcodec foo -acodec bar output
[08:50:43 CET] <Commander_Cody> How could I resume that if my computer crashed while running it?
[09:03:04 CET] <tdr> next time run it through a job scheduler with checkpoints
[09:05:15 CET] <Commander_Cody> tdr?
[12:04:13 CET] <faLUCE> Hello. I have file1.ts and file2.ts (video only, h264, mpegts).  both can be viewed with ffplay, vlc and mplayer, but file1 has some problems (timestamps?)  with vlc and mplayer . Here's the output of ffprobe for both:  http://paste.ubuntu.com/24108391/ .   Where can be the error on file1? I see that ffprobe shows framerate for file2 and no framerate for file1, but I did not set any framerate while creating both of them....
[12:04:14 CET] <faLUCE> .
[12:07:31 CET] <faLUCE> In addition, I see that ffprobe for file1 shows "1/2000000", which I don't understand
[13:09:38 CET] <oal> I'm trying to record a screencast with ffmpeg in Linux (Gnome 3) and I get flickering where sometimes the window behind the active window shines trhough
[13:11:38 CET] <oal> https://dl.dropboxusercontent.com/u/262587/output.avi like in the beginning of this short recording
[13:11:54 CET] <oal> Any ideas how to avoid that?
[14:44:45 CET] <faLUCE> Hello. When probing somefile.ts I obtain this info:  Stream #0:0, 61, 1/1200000: Video: h264 (Constrained Baseline), 1 reference frame, yuv420p(left), 640x480, 1/2000000, 25 fps, 1000k tbr, 1200k tbn, 2000k tbc  .   What is "1/2000000"? Is it the timebase of the encoder?
[16:32:20 CET] <Threads> any reason ffmpeg is bad at concating mkv but its fine with mp4 ?
[16:32:41 CET] <Threads> keep getting something about autoh264blahblah with mkv but its fine with mp4
[16:36:13 CET] <BtbN> probably because the contents of your mp4 files match, but the mkvs don't.
[17:36:05 CET] <thebombzen_> hey, I upgraded cuda (to 8.0 I think) and nvenc doesn't work anymore. is this known or is there something wrong with my setup?
[17:36:40 CET] <durandal_1707> your setup
[17:36:58 CET] <durandal_1707> dont touch something that works
[17:38:48 CET] <thebombzen_> well I'm sorry I'm on Arch and upgraded my system?
[17:39:19 CET] <thebombzen_> either way, what's interesting is that if I run: ffmpeg -f lavfi -i testsrc -c h264_nvenc -gpu list -f null -
[17:39:25 CET] <klaxa> i'm surprised my computer didn't catch fire when i updated my arch given how much shit updates broke
[17:39:45 CET] <thebombzen_> if I run that command, it'll hang for a bit, but then it'll dow hat's expected
[17:39:47 CET] <thebombzen_> list my two GPUs
[17:40:00 CET] <thebombzen_> and then exit immediately after listing both gpus
[17:40:40 CET] <thebombzen_> however, if I actually attempt to do a test encode (e.g. with -gpu any instead of -gpu list) then it'll hang and then fail
[17:40:47 CET] <thebombzen_> claiming that no nvenc-capable devices found.
[17:41:03 CET] <thebombzen_> lemme pastebin my uncut output one second
[17:42:58 CET] <thebombzen_> oh interesting, this is different behavior than yesterday
[17:44:19 CET] <thebombzen_> can anyone give me some insight on this error? "[h264_nvenc @ 0x55c547cb7860] Failed creating CUDA context for NVENC: 0x3e7"
[17:44:53 CET] <thebombzen_> now according to dur (the maintainer) it's not an ffmpeg issue and I believe that but does anyone have any insight into the problem?
[17:46:53 CET] <jkqxz> <http://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__TYPES.html#group__CUDA__TYPES_1gc6c391505e117393cc2558fff6bfc2e9> says that return code is "unknown internal error".
[17:47:03 CET] <jkqxz> So you probably need the source code for the nvidia driver to determine anything at all.
[17:48:48 CET] <jkqxz> If you don't have the source code then sucks to be you.  Maybe reinstall it, or try a different version?
[18:11:21 CET] <BtbN> thebombzen, "updated CUDA"? So, you updated your nvidia driver?
[18:11:28 CET] <BtbN> Because that's the only external thing ffmpeg uses.
[18:30:39 CET] <thebombzen_> upon some googling, it looks like there is a bug, where if you suspend your system after an nvenc session has happened, and then reload the system, you won't be able to try again
[18:30:51 CET] <thebombzen_> and reloading the nvidia_uvm module should fix this
[18:31:21 CET] <thebombzen_> unfortunately for me it didn't fix it, AND my system didn't get suspended, so I don't know what to try here
[18:32:37 CET] <BtbN> reboot
[18:32:48 CET] <BtbN> If it still doesn't work, something's wrong with your driver.
[18:34:49 CET] <thebombzen_> well I didn't upgrade the driver, just cuda. but yea reloading the driver is a thing. I just didn't want to "reboot fixes it" cause it feels like I'm on NT or something. but when we're talking drivers and kmods then it might actually be the right thing to do
[18:35:12 CET] <BtbN> There is nothing to upgrade there.
[18:35:37 CET] <BtbN> ffmpeg purely depends on the nvidia driver, except for libnpp, which is non-free though
[18:37:17 CET] <thebombzen_> well, it may be a surprise, but cuda isn't teh same thing as the nvidia driver
[18:37:33 CET] <thebombzen_> nvidia's driver is with the Arch package "nvidia" and cuda is with the package "cuda"
[18:38:22 CET] <thebombzen_> nvidia is the thing that installs the kernel module that allows your graphics card to render your display (i.e. The Driver) and the Cuda package provides the headers and libraries to just the Common Unified Device Architecture
[18:38:31 CET] <thebombzen_> so no, cuda != drivers
[18:39:14 CET] <JEEB> cuda being the cuda sdk then, because I'm pretty sure the runtime comes with the drivers (although I'm not 100% sure since I haven't run CUDA-using apps lately :P)
[18:39:18 CET] <BtbN> all libraries ffmpeg uses for nvenc and cuvid come from the nvidia driver
[18:39:37 CET] <BtbN> It has no dependencies on CUDA specific libraries(except for libnpp), and it has no dependencies on any external headers.
[18:40:17 CET] <thebombzen_> this is not correct
[18:40:18 CET] <thebombzen_> http://0x0.st/JR2.txt
[18:40:30 CET] <BtbN> those come from the nvidia-driver.
[18:40:40 CET] <thebombzen_> no they don't
[18:40:44 CET] <BtbN> yes they do
[18:40:48 CET] <thebombzen_> the nvidia driver doesn't have libcuda, I'm trying to tell you
[18:40:54 CET] <BtbN> It does
[18:40:59 CET] <thebombzen_> literally click the link
[18:41:12 CET] <BtbN> I did, and those libraries are part of the nvidia driver.
[18:41:27 CET] <JEEB> I've not chekced the actual nvidia archives but it's quite possible that arch has just separated things around in addition to the SDK :P
[18:41:44 CET] <thebombzen_> well then if you clicked the link
[18:41:44 CET] <JEEB> the SDK *is* definitely separate from the driver
[18:41:55 CET] <thebombzen_> and they're part of the Nvidia driver, then why are they not in the "nvidia" package
[18:42:01 CET] <thebombzen_> which doesn't depend on cuda
[18:42:03 CET] <JEEB> because arch?
[18:42:16 CET] <thebombzen_> yea but it's possible to install the nvidia driver without installing libcuda
[18:42:18 CET] <JEEB> although like I just said I haven't looked at the contents of the latest nvidia binaries
[18:42:23 CET] <thebombzen_> ergo they're NOT part of the nvidia driver
[18:42:24 CET] <BtbN> Only if you use some botched package
[18:42:34 CET] <BtbN> The official installer and any sane package of it brings it
[18:42:39 CET] <thebombzen_> BtbN: do you just not know when to quit
[18:42:59 CET] <JEEB> umm, what BtbN seems to be saying is that arch took the thing that nvidia distributes
[18:43:02 CET] <JEEB> and separated it
[18:43:05 CET] <JEEB> into two packages
[18:43:16 CET] <thebombzen_> yea, but the kernel module doesn't actually pull in cuda
[18:43:18 CET] <BtbN> I'm looking at the nvidia driver archive right now, and it's all there.
[18:43:28 CET] <thebombzen_> you can install the driver without installing cuda
[18:43:30 CET] <JEEB> thebombzen_: that's something arch decided
[18:43:40 CET] <BtbN> There is no "Pull in CUDA". The CUDA SDK is just for development against it. Everything else for the runtime comes with the driver.
[18:43:53 CET] <BtbN> And ffmpeg does not need the SDK for building.
[18:43:56 CET] <JEEB> I'd trust on BtbN what nvidia distributes within the driver package
[18:44:13 CET] <thebombzen_> BtbN: it's not accurate that the runtime comes with the driver.
[18:44:15 CET] <JEEB> arch can optimize and separate and cut things around how it wants, I guess
[18:44:21 CET] <BtbN> Yes it is...
[18:44:27 CET] <thebombzen_> libcuda.so comes with cuda, not with nvidia
[18:44:27 CET] <JEEB> thebombzen_: for your distribution, because they decided to cut it around
[18:44:33 CET] <thebombzen_> YES FOR MY DISTRIBUTION. THIS MATTERS.
[18:44:44 CET] <thebombzen_> because my statement was "I updated cuda but didn't update my driver"
[18:44:58 CET] <JEEB> ok, as long as you understand. but for anyone else the official driver package is what consists as a driver
[18:44:58 CET] <thebombzen_> BtbN is trying to claim that this is impossible because they're distributed together
[18:45:09 CET] <BtbN> Well, if you mix different versions of files that are supposed to be installed together with one driver installer, stuff exploding is about to be expected.
[18:45:12 CET] <thebombzen_> and I'm saying, no it's definitely possible, because I literally just did it.
[18:45:37 CET] <thebombzen_> so don't try to tell me that I can't upgrade cuda without upgrading the driver, because I just did, and I can send you the file listing of the separate packages if you don't believe me.
[18:45:49 CET] <BtbN> Nvidia gives absolutely no backward or forward compatiblity between their driver releases and libraries from them
[18:46:02 CET] <BtbN> Seems to be working great what you just did
[18:46:23 CET] <thebombzen_> http://sprunge.us/FOZX
[18:46:24 CET] <BtbN> And I can send you the file listing of all files coming from x11-drivers/nvidia-drivers, so?
[18:46:47 CET] <thebombzen_> on your distribution? yea but that doesn't matter
[18:46:47 CET] <BtbN> Doesn't change anything. All that matters is how the nvidia driver is intended to be used. If Arch decided to split it, you have to live with the consequences of that.
[18:47:05 CET] <BtbN> So your distribution matters, but none else. Great.
[18:47:05 CET] <thebombzen_> I'm only trying to explain to you is that I did actually udpate cuda without updating the driver.
[18:47:21 CET] <thebombzen_> well you're the one who said, "You can't update cuda without updating the driver."
[18:47:22 CET] <BtbN> And I'm trying to explain you that that's not intended and you already found out what it causes.
[18:47:24 CET] <thebombzen_> Well, yes, I can.
[18:50:11 CET] <thebombzen> well either way, I'm going to reboot. I don't like that nvidia makes you do that all the time but they're probably not used to catering to users who usually don't have to do that
[18:51:15 CET] <BtbN> unloading all (nvidia) kernel modules and completely restarting X should have the same effect
[18:51:40 CET] <BtbN> And if you really have mixed versions of the cuda libraies and the rest of the nvidia driver: fix that
[18:55:15 CET] <thebombzen_> either way, just reloading the kernel module by rebooting fixed it. I just find it strange that it would be necessary
[18:55:31 CET] <thebombzen_> given that I tried unloading and reloading the module and that didn't do it. (rmmod && modprobe)
[19:00:10 CET] <BtbN> all modules?
[19:00:12 CET] <BtbN> Or just uvm?
[19:00:21 CET] <BtbN> Nvidia has quite a few by now
[19:00:25 CET] <BtbN> uvm, kvm, nvidia, ...
[19:00:31 CET] <BtbN> *kms
[19:12:45 CET] <faLUCE> Hello. I have file1.ts and file2.ts (video only, h264, mpegts).  both can be viewed with ffplay and vlc, but file1 is not seekable with vlc. Is there a way to detect if it contains errors, with ffprobe? Otherwise, what could I check?
[19:14:34 CET] <faLUCE> I compared them with wireshark, and the only difference seems "random access indicator" flag, which is set to 0 in file1, and to 1 in file2
[19:36:35 CET] <thebombzen_> the module in question was uvm
[19:36:40 CET] <thebombzen_> that's teh one that handles the nvenc
[19:36:54 CET] <thebombzen_> the module "nvidia" just drives the display
[19:37:58 CET] <thebombzen_> I have loaded nvidia, nvidia_drm, nvidia_uvm, and nvidia_modeset
[19:44:03 CET] <nightlingo> hello!
[19:44:14 CET] <thebombzen_> Hello
[19:54:27 CET] <nightlingo> I want to split a small portion of a .mp4 video using ffmpeg, without re-encoding it and get the output sent to stdout. I reallized that ffmpeg doesn't support outputting an .mp4 video to stdout because the MP4 muxer needs the ability to seek back in a stream. Then I tried outputting a fragmented mp4 video, which worked for stdout, but it turns out it is not supported by several players like Windows 7 Media Player. Any other suggestion
[19:54:28 CET] <nightlingo> please?
[19:56:39 CET] <DHE> faLUCE: RAI should be set on any video keyframe
[19:56:49 CET] <DHE> or at least the first segment of i
[19:56:50 CET] <DHE> it
[20:38:37 CET] <BtbN> thebombzen_, you still have to reload all of them. They all interact
[20:38:51 CET] <BtbN> Might even get a version mismatch message in dmesg if you don't.
[20:48:49 CET] <thebombzen_> well rebooting fixed it
[20:49:05 CET] <thebombzen_> but the solution I found specifically said "reload nvidia_uvm"
[20:49:38 CET] <thebombzen_> doing "rmmod nvidia" would probably have messed it up a bit more because I'd lose my xserver, which would be sort of a pain because my display manager is on
[20:50:37 CET] <thebombzen_> if I shut off lightdm then upon restarting I'd have a graphical prompt to log in to a user account, which is not possible for me to do easily remotely
[20:50:49 CET] <thebombzen_> (I'm operating over ssh/vnc)
[20:51:08 CET] <thebombzen_> and I'd lose graphical vnc access if necessary
[20:51:30 CET] <thebombzen_> rebooting doesn't have the same issue because I use PAM to do the initial login so I bypass that problem.
[20:52:58 CET] <BtbN> yes, an X restart is required as well
[21:04:16 CET] <thebombzen_> well restarting X is best done as rebooting
[21:04:21 CET] <thebombzen_> because I use lightdm
[21:04:43 CET] <furq> doesn't restarting lightdm restart x
[21:05:16 CET] <johnnie> can you tell me how to install as a depedency in FF Multi Convert please?
[21:05:26 CET] <durandal_1707> no
[21:17:58 CET] <SpeakerToMeat> When outputing to discrete frame files, with this syntax: "Filename_%06d.dpx" is there any way to make the index for the number start at possition X ?
[21:18:58 CET] <jkqxz> -start_number X
[21:19:17 CET] <SpeakerToMeat> thanks jkqxz
[21:51:11 CET] <faLUCE> DHE: thnks
[22:08:12 CET] <faLUCE> DHE and others: is it possible to get the keyframe/not keyframe info from an AVPacket struct? Should I check the "flags" field?
[22:31:23 CET] <madguy02_> hii, i wanted to know,how do i get started with ffmpeg?
[22:37:38 CET] <llamapixel> madguy02_: google ffmpeg basics
[22:37:58 CET] <llamapixel> the keycorner.org link isnt bad
[22:47:10 CET] <grym> llamapixel: what a weird place to put ffmpeg docs
[23:16:17 CET] <_jam> is there a way to get ffmpeg to print the libmp3lame version?
[23:16:54 CET] <_jam> -version does not include libmp3lame in the list of library versions printed
[23:21:23 CET] <DHE> if it's a dynamic link against mp3lame, you can check with `ldd` to see what it's linked against and check it
[23:33:28 CET] <_jam> on my debian system, i know how to do that with apt, but i was hoping for a more robust method
[23:33:48 CET] <_jam> platform independent and all
[00:00:00 CET] --- Sun Mar  5 2017