[Ffmpeg-devel-irc] ffmpeg.log.20171022

[burek]

[07:48:18 CEST] <Srini> hello i am looking for downloading the dash live segments
[07:49:30 CEST] <Srini> and when i run the ffmpeg command it downloads one recent segment and quits, is there any way i can continue to download the segments the way it is done in hls m3u8?
[07:52:03 CEST] <Srini> i am using the base url in my ffmpeg command
[08:16:03 CEST] <Srini> DHE: ^^
[10:48:40 CEST] <johnny_|_> Hi. I am using ffmpeg.js. Is it possible to input live stream (h.264) and get transcoded (vp8) stream on the output?
[11:43:26 CEST] <blap> it's 2017 and we don't have auto voice translation of movies yet
[14:42:43 CEST] <Pandela> Has anyone had any luck with piped svg sequences? 3.4 says it has an svg_pipe format now, but piping in svg's just makes ffplay/ffmpeg hang
[15:31:12 CEST] <Gaulois94> Hello
[15:31:56 CEST] <Gaulois94> I have few question about ffmpeg and RTP video stream. I am reading a .sdp file using ffmpeg in C/C++, and my question is about reading : how do you manage to handle the time synchronization ?
[15:32:58 CEST] <Gaulois94> WIll av_read_frame be in pause while there is no images ?
[15:34:02 CEST] <DHE> av_read_frame may block on network activity, but as long as sufficient data is available it will return the next AVPacket to you
[15:34:19 CEST] <DHE> the ffmpeg CLI tool uses all these same APIs
[15:34:43 CEST] <Gaulois94> Ok, I don't have in the client size handle the time synchronization, av_read_frame does it for me
[15:34:59 CEST] <Gaulois94> have to*
[15:36:16 CEST] <DHE> no it doesn't. if you're using the network, you're just experiencing the flow of data from the remote side
[15:36:32 CEST] <jamesl> I'm trying to stream the output of a uv4l webcam (/dev/video0) to Twitch.tv . I am using this command: https://pastebin.com/raw/TciAbst1 and it sends packets successfully, but the Twitch.tv stream is offline. What is causing this?
[15:36:56 CEST] <Gaulois94> Ok, that is good enough for me
[15:37:16 CEST] <Gaulois94> (I do the synchronization in the remote side), thank you
[15:37:32 CEST] <DHE> unless the network is highly trusted (ie. local LAN) that's still not considered good enough
[15:37:43 CEST] <DHE> there should be at least some buffering on the client side
[15:38:32 CEST] <Gaulois94> It is indeed in local LAN
[15:39:26 CEST] <jamesl> When I change the stream key it goes from "offline" to "authentication failed" so I know the twitch URL and stream key are correct
[15:39:45 CEST] <jamesl> https://trac.ffmpeg.org/wiki/StreamingGuide I followed this but it's for x11grab, not v4l
[16:25:30 CEST] <JnvSor> Is there a way to set which audio/subtitle track is default? I can do it with -map and remapping all the tracks, but some files have upwards of 20 streams and then I have to map them all individually... I'd rather just say "Mark this stream as default"
[16:26:53 CEST] <JEEB> if it's the actual selection that you're not OK with then I think map is the only way to do it with ffmpeg.c, which goes into manual mode and thus requires you to map everything you need
[16:27:03 CEST] <JEEB> because tehre is a separate metadata of "default track" in some containers :P
[16:27:15 CEST] <JEEB> which you might have also been meaning
[16:27:21 CEST] <JEEB> "I want X to be the default one in the output"
[16:28:06 CEST] <JnvSor> Yeah that's what I mean. Does mkv container have a way to mark a stream as default for output?
[16:29:41 CEST] <JEEB> there's a default flag yes
[16:29:53 CEST] <JEEB> no idea how you set it in ffmpeg.c
[16:29:55 CEST] <JEEB> vOv
[16:31:42 CEST] <JEEB> ok, it was in the examples for the disposition parameter
[16:31:45 CEST] <JEEB> "-disposition:a:1 default"
[16:31:55 CEST] <JEEB> would set disposition for the second audio stream to 'default'
[16:32:26 CEST] <JEEB> see https://www.ffmpeg.org/ffmpeg-all.html
[16:37:38 CEST] <jamesl> I'm trying to stream the output of a uv4l webcam (/dev/video0) to Twitch.tv . I am using this command: https://pastebin.com/raw/TciAbst1 and it sends packets successfully, but the Twitch.tv stream is offline. What is causing this?
[16:42:58 CEST] <JnvSor> I read that and it didn't work... Just realized I have ffmpeg pointing to a custom build I set up years ago... Yay... Stupid me, never mind :/
[17:11:07 CEST] <beastd> JnvSor,JEEB: I think setting default is not really controlable via ffmpeg. I switched to using mkvmerge for the task of marking the default track.
[17:12:45 CEST] <beastd> Because unfortunately it wasn't trivial to add functionality to ffmpeg. Some time has passed and I forget the details... Though I think i can at least dig out the mkgmerge syntax I used...
[17:23:48 CEST] <jamesl> Why would a ffmpeg command work with x11grab but not with a uv4l webcam? Here's the command:  https://pastebin.com/raw/TciAbst1
[17:26:05 CEST] <JnvSor> beastd: Worked for me, just needed to mark the first track disposition none as well
[17:27:31 CEST] <beastd> JnvSor: Ah ok. That is great to hear. I don't know why it didn't work for me back then...
[17:39:24 CEST] <furq> beastd: i'm pretty sure that didn't work until relatively recently
[18:47:51 CEST] <srini> logger url
[19:12:28 CEST] <jamesl> Why would a ffmpeg command work with x11grab but not with a uv4l webcam? Here's the command:  https://pastebin.com/raw/TciAbst1
[19:14:12 CEST] <c_14> what's the error?
[19:18:18 CEST] <jamesl> c_14: the stream appears to be blank
[19:18:35 CEST] <jamesl> the stream to Twitch works when it used x11grab
[19:19:13 CEST] <c_14> are you sure that v4l2 source input is working?
[19:19:27 CEST] <jamesl> yeah, I recorded it to a file
[19:19:31 CEST] <jamesl> the webcam works
[19:19:32 CEST] <c_14> and that works?
[19:19:57 CEST] <c_14> maybe the framerate/gop size isn't stable
[19:20:06 CEST] <c_14> or the bitrate is wrong
[19:20:14 CEST] <c_14> twitch has a bunch of things it cares about
[19:20:50 CEST] <jamesl> the bitrate is 95.4 kbit/s and is reasonably stable
[19:23:22 CEST] <BtbN> streaming something that low to Twitch might just not be supported
[21:42:10 CEST] <Gaulois94> Hi, do you know why sws_getContext send the signal "abort" ? It stopped if my debugger is correct at sws_setColorspaceDetails
[21:43:07 CEST] <Gaulois94> Or maybe I am stupid and I didn't checked if my codec was correct
[22:01:00 CEST] <dystopia_> is there someway to convert latm aac headers to adts without re encoding the file
[22:45:15 CEST] <DHE> dystopia_: if there is, it'll be a bitstream filter (ffmpeg -bsfs)
[22:46:43 CEST] <furq> there's an adts to asc bsf, but i don't see the inverse
[22:47:34 CEST] <furq> actually nvm i guess latm is something different
[22:48:46 CEST] <furq> dystopia_: there's a -latm 1 switch, so i guess -c copy out.aac with no -latm set will do it?
[22:48:56 CEST] <furq> not checked though
[23:01:09 CEST] <long-klong> hello !
[23:01:24 CEST] <long-klong> I am a student who is working on a super resolution algorithm
[23:01:43 CEST] <long-klong> and I am fidgeting around with the motion estimation part
[23:02:10 CEST] <JEEB> you might want to talk to hanna I guess?
[23:02:19 CEST] <long-klong> is familiar with the following papers ?
[23:02:59 CEST] <hanna> I'm not that familiar with the subject either
[23:03:05 CEST] <hanna> but I'd be interested in implementing this sort of thing in GLSL
[23:03:26 CEST] <JEEB> that's mostly why I noted, since CPU implementations tend to be zomg zufall slow
[23:03:59 CEST] <long-klong> https://www.researchgate.net/profile/Horst_Bischof/publication/248964741_A_Duality_Based_Approach_for_Realtime_TV-L1_Optical_Flow/links/0fcfd51113a883c09a000000.pdf?inViewer=true&disableCoverPage=true&origin=publication_detail
[23:04:35 CEST] <long-klong> https://vision.in.tum.de/_media/spezial/bib/mitzel_et_al_dagm09.pdf
[23:04:49 CEST] <blap> i approve of this nick
[23:05:21 CEST] <long-klong> http://www.caam.rice.edu/~zhang/caam699/opt-flow/horn81.pdf
[23:05:25 CEST] <long-klong> these are the papers
[23:06:48 CEST] <long-klong> so if there is someone who can help me with the optical flow.. I will be grateful
[23:07:12 CEST] <JEEB> oh, I thought you had something kind of working :D
[23:08:08 CEST] <hanna> I've never looked at optical flow
[23:08:15 CEST] <hanna> The only methods I'm familiar with are block estimation and convolution
[23:08:40 CEST] <hanna> doesn't optical flow have issues due to lack of training data and occlusion etc.?
[23:10:50 CEST] <long-klong> I do not know why I got signed out from the network
[23:10:54 CEST] <long-klong> I am still here
[23:12:13 CEST] <JEEB> http://up-cat.net/p/bb86140d
[23:12:15 CEST] <JEEB> long-klong: ^
[23:13:14 CEST] <long-klong> thank you JEEB
[23:13:16 CEST] <long-klong> !
[23:13:18 CEST] <long-klong> mised that part
[23:13:28 CEST] <long-klong> I do not know
[23:13:35 CEST] <long-klong> I am just a sstudent hanna
[23:13:45 CEST] <long-klong> can you elaborate please on what you know ?
[23:13:58 CEST] <hanna> I don't know much more
[23:14:01 CEST] <hanna> I am also just a student
[23:14:06 CEST] <long-klong> I read something related to occlusion
[23:15:12 CEST] <long-klong> did you do convolution of images or signals and systems stuf ?
[23:35:00 CEST] <hanna> I read a paper about doing motion estimation using convolutional neural networks
[23:35:20 CEST] <hanna> which handles occlusion very well and doesn't have any issues with training data
[23:35:26 CEST] <hanna> but it's too slow for realtime
[23:35:44 CEST] <long-klong> hmm... cool
[23:36:40 CEST] <long-klong> I am interested in copmputer vision at the moment...maybe later in Artificial intelligence (PhD ?) .
[23:37:29 CEST] <long-klong> I am an electronics and telecommunications student that wants to do an image processing algorithm for his bachelor thesis , implemented on an FPGA
[23:37:39 CEST] <long-klong> :)
[23:38:07 CEST] <long-klong> I am going to die probably in the next 6 months
[23:38:09 CEST] <long-klong> lol
[23:43:37 CEST] <long-klong> hey hanna ... what was block estimation ?
[23:44:06 CEST] <hanna> long-klong: block estimation is what MVTools, SVP etc. do
[23:44:16 CEST] <hanna> basically block motion search (like what encoders do)
[23:44:24 CEST] <hanna> + algorithms to do pixel reflowing based on the motion vectors
[23:47:25 CEST] <thebombzen> hanna: how is that better than like
[23:47:30 CEST] <thebombzen> exhaustive searching
[23:47:50 CEST] <hanna> exhaustive search is an example of a block motion search strategy
[23:48:09 CEST] <thebombzen> oh you mean this doesn't use MBs to estimate motion vectors
[23:48:17 CEST] <thebombzen> sounds cool
[23:59:21 CEST] <hanna> MBs?
[00:00:00 CEST] --- Mon Oct 23 2017