[Ffmpeg-devel-irc] ffmpeg.log.20190807

[burek]

[00:06:35 CEST] <TheAMM> Can someone quickly elaborate what the "GIF parser" in 4.2 is about
[00:06:45 CEST] <TheAMM> I'm not seeing any recent major changes in gifdec
[00:08:57 CEST] <BtbN> It's probably about faca28c264bee8ff93637ae69b62d5e974eee8a8..ec8502f9cb96a83704c45e94d11f2434253f9231
[00:10:12 CEST] <BtbN> if in doubt git blame the changelog
[00:10:49 CEST] <furq> https://github.com/FFmpeg/FFmpeg/commits/master/libavcodec/gif.c
[00:11:48 CEST] <furq> is that support for per-frame palettes i see
[00:12:56 CEST] <TheAMM> hmh
[00:13:02 CEST] <TheAMM> Alright, thanks
[00:21:56 CEST] <durandal_1707> see lavc/gif_parser.c
[04:50:55 CEST] <Dotz0cat> i have to say thanks what yall said got it working
[04:51:25 CEST] <Dotz0cat> now all i have to worry about is the rain clouds and lighting
[11:54:26 CEST] <Spring> are there any docs I could check out about the 'GIF parser' mentioned in the ffmpeg 4.2 changelog?
[11:54:49 CEST] <Spring> as in, what has been added/changed with GIF support with this release
[12:22:16 CEST] <another> relaxed: https://johnvansickle.com/ffmpeg/release-readme.txt could use some update
[12:55:32 CEST] <durandal_1707> Spring: gif support was removed due patents
[13:06:59 CEST] <Spring> durandal_1707, ? The patents expired years ago last I read
[14:18:04 CEST] <rswarbrick> Hi all! A question about the -shortest flag. I'm running a build from master (N-94482-g6c67c8ca9a). I want to use the libvmaf binding to compare two videos with something like 'ffmpeg -i A.y4m -i B.y4m -lavfi "[0:v][1:v]libvmaf" -f null -'. This works if A and B are the same length, but gives low numbers if (say) A is shorter. Adding -shortest to the command doesn't seem to have any effect if I also use the -f flag. Is this by design?
[14:18:04 CEST] <rswarbrick> If so, how to I write out a null video and also use the shorter of the two inputs?
[14:20:16 CEST] <kepstin> -shortest works by pulling all the videos through the filters until one returns an eof, so the filters will see a little past the point where the final video stops
[14:21:00 CEST] <kepstin> i don't know any way to do what you want other than check the video lengths in advance and trim them before the filter
[14:21:02 CEST] <rswarbrick> Incidentally, this command sort of works: "ffmpeg -i A.y4m -i B.y4m -lavfi "[0:v][1:v]libvmaf" -shortest foo.y4m", except that it writes a zero-length file and gives a "conversion failed" error message.
[14:21:30 CEST] <rswarbrick> kepstin: Hmm, I see. So the command above might be passing some other random rubbish to libvmaf for the start of the N+1'st frame?
[14:21:46 CEST] <kepstin> no, it passes an end of file indicator to the filter.
[14:21:54 CEST] <kepstin> i dunno what the filter itself does with that
[14:22:04 CEST] <kepstin> (might be configurable)
[14:22:20 CEST] <rswarbrick> Oh, so libvmaf will get "frame, frame, frame, EOF, some-rubbish"?
[14:22:27 CEST] <rswarbrick> Or "frame, frame, frame, some-rubbish, EOF"?
[14:22:40 CEST] <kepstin> ah, vmaf filter uses framesync options: https://www.ffmpeg.org/ffmpeg-filters.html#framesync
[14:22:48 CEST] <kepstin> default on eof is it repeats the last seen frame
[14:22:52 CEST] <kepstin> you should change that
[14:22:54 CEST] <rswarbrick> Ahah!
[14:23:41 CEST] <rswarbrick> Hmm. Now I have to work out the syntax... :-/
[14:24:06 CEST] <kepstin> -filter_complex [0:v][1:v]libvmaf=shortest=1
[14:24:17 CEST] <kepstin> but do read the filter syntax documentation.
[14:25:26 CEST] <rswarbrick> Ah, thanks. So that means: "There is just one node in the filter graph, using libvmaf. It takes video from input streams 0 and 1. Also, it sohuld be configured by setting "shortest" to 1".?
[14:25:34 CEST] <rswarbrick> *shoult
[14:25:38 CEST] <rswarbrick> **should.
[14:25:40 CEST] <rswarbrick> Oh dear.
[14:28:25 CEST] <rswarbrick> Brilliant, it works! Thank you very much for the help.
[14:32:08 CEST] <Freneticks> I try to copy a stream in mp4 format, with segmenter or hls segmenter but the output file seems always corrupt : ffprobe say : Invalid data found when processing input
[14:32:25 CEST] <Freneticks> Of course i wait the segment is finished
[14:32:53 CEST] <Freneticks> What I'm missing ?
[15:53:44 CEST] <Freneticks> Maybe I found my problem : is mpeg-ts can be remuxed in pure h264 ? or need to be rencoded ?
[15:56:17 CEST] <DHE> you may need -bsf:v h264_mp4toannexb
[18:57:38 CEST] <Saccarab> what is the best way to batch merge audio files where each file has a specific start timestamp
[18:57:47 CEST] <Saccarab> adelay works great but I would rather not re encode
[19:04:03 CEST] <Saccarab> I've also tried demux concat and added silence between each file instead of start timestamps but that after some count of files I get synchronization issues
[19:34:39 CEST] <relaxed> another: updated, thanks
[20:24:14 CEST] <julesverne> .bet 12%
[20:25:06 CEST] <julesverne> sorry. wrong chat
[22:13:47 CEST] <ossifrage> I'm doing some experiments with trying to send video over websockets and I was wondering if someone could recommend a Media Source Extension player that is both hackable and not excessively bloated
[22:15:21 CEST] <ossifrage> I gave up on every getting reasonable latency out of dash/hls, but I'm hoping that maybe some websocket horror would be able to approach what I get with rtsp
[22:31:33 CEST] <ocrete> ossifrage: if you want really low latency (video call grade), then you need to go webrtc instead...
[22:32:47 CEST] <ossifrage> ocrete, I'm looking for 300-500ms (which is what I get from rtsp over both udp and tcp)
[22:32:55 CEST] <ocrete> yeah then that's webrtc..
[22:33:36 CEST] <ossifrage> I'm going to see what I get with websockets first before having to deal with the extra complexity of webrtc
[22:58:01 CEST] <ossifrage> I can get ~1GOP of latency just serving an infinite mp4 file over http, but that sucks because it is bursty
[00:00:00 CEST] --- Thu Aug  8 2019