[Ffmpeg-devel-irc] ffmpeg.log.20171111

[burek]

[01:30:38 CET] <SortaCore> does libx264 have a decoder?
[01:37:06 CET] <DHE> no, it is just an encoder. ffmpeg includes an h264 decoder built-in
[01:37:28 CET] <DHE> internal decoder, external/3rd party decoder (with ffmpeg API)
[01:37:54 CET] <potatopants> I've got a H264 MP4 video I'm working with - it's already quite compressed so I want to avoid transcoding. I can achieve what I want by making a series of cuts and joins, with the exception of one fade-to-black transition which I plan to extract, transcode and rejoin. Do I need to do anything special with regard to I-frames etc or should I be OK making regular selections?
[01:39:00 CET] <DHE> if you're not transcoding, you must always start with a keyframe from the source material in your cuts
[01:43:00 CET] <potatopants> DHE, thanks. will ffmpeg do that for me automatically if i use -ss and -t? if not, how should I enumerate/select the right keyframes?
[01:43:55 CET] <DHE> depends if you specify it on the input or the output. the input will try to lock onto keyframes
[01:44:21 CET] <DHE> another idea is to use the segment or hls muxers with a really short time. the muxer will only cut on keyframes so you'll get short videos in keyframe segments... maybe that helps?
[01:44:31 CET] <DHE> hls is nice because the files can literally be concatenated together safely
[01:44:42 CET] <DHE> but the codec options are more restrictive
[01:44:56 CET] <SortaCore> any way to improve the decoding speed with HWA?
[01:45:03 CET] <SortaCore> using built-in H264 decoder
[01:45:20 CET] <furq> the builtin h264 decoder is software
[01:45:38 CET] <DHE> there are options for hw accel decoding, like vdpau. but they require some 3rd party drivers
[01:45:48 CET] <DHE> actually vdpau may be a bad example
[01:46:01 CET] <potatopants> DHE, thanks, I'll check out HLS. I was looking at segment also and will read into selecting on the input vs output.
[02:50:33 CET] <potatopants> I am trying to concat 2 mp4 files using the concat mixer - both of them came from the same source file, one of them has had a fade transition applied using -vf. both the files play fine individually, but when I concat at the union I get lots of green frames and frozen motion. I have pasted media details here: https://pastebin.com/KFip3C0c - have I made my files incompatible with the transcode i performed on one?
[03:39:14 CET] <potatopants> I managed to resolve the issue by first converting my mp4 files to mpeg transport streams and concatenating those :-)
[06:36:59 CET] <bledo> hls_flag delete_segments is a scam!
[06:37:03 CET] <bledo> i say.
[07:34:38 CET] <Johnjay> hmm i guess ffmpeg doesn't use much in the way of opengl does it?
[09:40:51 CET] <Guest86059> Hello everybody! I'm trying to add a still image to audio track to make it video :) Is it possible to reduce "video" track size close to the source image size to keep result size close to the source audiofile?
[09:41:50 CET] <Guest86059> The most way I found is here: https://superuser.com/questions/1041816/combine-one-image-one-audio-file-to-make-one-video-using-ffmpeg, -c:v libx264 -tune stillimage -pix_fmt yuv420p
[09:42:58 CET] <c_14> there's a few tricks you can do, increase the gop size (-g <large_number>), decrease the framerate (beware of decreasing it too much as many video players don't like that, increase the crf (decrease the bitrate)
[09:43:19 CET] <c_14> use a slower x264 preset
[09:43:54 CET] <Guest86059> This make a videotrack about the similar size as audio, so in this case result is 2x file size. Is it possible to reduce it even more, because the image actually is still, so should be no need to insert this image and deltas many times...
[09:44:42 CET] <Guest86059> or is the video format simple like wav - keyframes, frames (deltas), and nothing can be more smart organized?
[09:47:44 CET] <Guest86059> Oh yeah, there are a lot of loses actually... big gop will not let scroll at the exact position You want, very-very low framerate - I'm also curious, what will happens :) high crf can just degrade an image quality... But if there is no other way - well, then there is no other way :(
[09:48:49 CET] <c_14> the problem is that most video players will only seek to keyframes (at least in the default mode) and keyframes are big
[09:49:07 CET] <c_14> even those players that let you seek to arbitrary points, still snap to the nearest frame
[09:49:09 CET] <c_14> so
[09:50:01 CET] <Guest86059> Yeah, and if I upload such video to Youtube, then it will covert it with low gop and size will grow dramatically again
[09:51:30 CET] <Guest86059> I noticed, some players do like this: If You click between keyframes (with gop like 12 seconds) then You will see artifacts for the next 6 seconds, but the sound will be ok... Maybe this is a solution in my case...
[09:55:22 CET] <Guest86059> But some of the players allows scroll to the keyframe only :( This is a big problem...
[09:59:57 CET] <Guest86059> So thank You very much :)
[10:04:46 CET] <kingsley> Please show me a working example of using the frei0r_src source.
[14:44:09 CET] <dongs> how do i specify reference frames with nvenc/ffmpeg
[14:44:18 CET] <dongs> -ref is no longer a valid option in nvenc apparently
[14:44:25 CET] <dongs> and resulting files are refframes:1 always
[14:44:47 CET] <dongs> http://developer.download.nvidia.com/compute/redist/ffmpeg/1511-patch/FFMPEG-with-NVIDIA-Acceleration-on-Ubuntu_UG_v01.pdf (using on windows, but this doc references -ref which doesnt work
[14:45:16 CET] <dongs> doc from 2015 not working with opensauce software, what a fucking shock
[14:46:19 CET] <dongs> https://ffmpeg.org/pipermail/ffmpeg-devel/2015-January/167866.html
[14:46:21 CET] <dongs> maybe related
[14:59:01 CET] <BtbN> dongs, I don't see any reason why -refs wouldn't work. Only the default was semi-recently changed from 1, to 0, which means automatic.
[14:59:27 CET] <dongs> yeah actually it does. that doc said "ref" not "refs". and looks liek 4 is max on nvenc
[14:59:43 CET] <dongs> thats wehre i got confused (ref->refs and then a couple tries to see max
[14:59:49 CET] <BtbN> It just passes it through untouched
[14:59:56 CET] <BtbN> whatever nvenc does with it happens internally of the driver
[16:15:46 CET] <arpu> how does the vsync cfr  as input works ?
[16:16:00 CET] <arpu> do not get a stable framerate :/
[16:16:21 CET] <arpu> input is from a icecast  or rtmp server
[16:19:46 CET] <arpu> the problem is for chuncking to dash files i need a constant framerate
[16:30:00 CET] <blap> dynaudnorm defaults give me bad distortion on music, i found much better is: dynaudnorm=g=31:f=500:r=0.8:p=0.90:m=2
[16:31:14 CET] <blap> if anyone has found their own optimum paramaters i'd be curious to see them
[16:31:45 CET] <blap> ping me if you have a reply
[18:09:54 CET] <ErAzOr> Hi together! I try to add a logo to my stream which is working fine. But the problem is, that the transparent background is actually not transparent in the final result. The background, which should be transparent, is actually white. Can anyone point me to the right direction?
[18:10:02 CET] <ErAzOr> this is the command I'm using: https://pastebin.com/AKjsAxiy
[18:36:46 CET] <TikityTik> Is there a way I can set the durection of a cut without using duration? I want to cut by the end timestamp
[18:48:30 CET] <TikityTik> nevermind, it's -to
[18:48:38 CET] <TikityTik> though my older version of ffmpeg didn't work for that
[18:57:58 CET] <ZeroWalker> does ffplay exist on android?
[18:59:39 CET] <JEEB> ZeroWalker: no, I recommend you utilize one of the actual players' libraries on android: either libvlc or libmpv
[18:59:54 CET] <DHE> ErAzOr: does the PNG actually contain transparency? eg: in GIMP the image would have a gray checkerboard background rather than solid white
[19:03:38 CET] <TikityTik> when you are doing 2-pass with ffmpeg, where does the pass data get stored?
[19:06:16 CET] <JEEB> depends on the encoder, they usually create a log file
[19:06:24 CET] <JEEB> usually in your current working directory
[19:07:03 CET] <DHE> you can also specify with -passlogfile
[19:07:28 CET] <DHE> depending on the codec and settings, more than one file may be needed in which case the passed parameter is treated as a filename prefix
[19:09:42 CET] <ZeroWalker> do they support the "tcp ?listen" thingy, if so that's awesome:)
[20:23:06 CET] <SortaCore> why are there a h264_nvenc, nvenc_h264 and nvenc encoders?
[20:30:30 CET] <DHE> because new cards support h264 and h265 codecs.
[20:30:41 CET] <DHE> oh, whoops that's not that
[20:30:56 CET] <DHE> they're compatibility names because we've had a few different names and have tried to standardize
[20:32:07 CET] <SortaCore> hmm alright, so which is the one to use for nvidia powered h264 encoding
[20:33:14 CET] <DHE> h264_nvenc is now preferred (codec, then method)
[20:39:42 CET] <SortaCore> alrighty
[21:27:17 CET] <BtbN> basically the one that does not throw a big deprecation warning
[22:58:41 CET] <kingsley> Do you happen to know how to use the frei0r_src source?
[23:00:04 CET] <durandal_1707> like any source video filter
[23:00:39 CET] <kingsley> It's documented at https://ffmpeg.org/ffmpeg-all.html#frei0r_005fsrc
[23:00:59 CET] <durandal_1707> ffmpeg -f lavfi -i frei0r_src=... ...
[23:01:15 CET] <kingsley> However, I've yet to actually make it work.
[23:01:28 CET] <durandal_1707> what error you get?
[23:01:42 CET] <kingsley> durandal_1707: I've love to see a complete working example.
[23:01:52 CET] <kingsley> s/I've/I'd/
[23:03:12 CET] <durandal_1707> kingsley: just put after = whats in documentation but without overlay
[23:04:04 CET] <durandal_1707> if its too complicated for you to grasp syntax perhaps you should reconsider your priorities
[23:04:13 CET] <kingsley> durandal_1707: OK, thank you.
[23:04:41 CET] <kingsley> I'm happy to report that your fine suggestion resulted in a nice working video.
[23:12:47 CET] <kingsley> For what it's worth, the following syntax works for me...
[23:12:50 CET] <kingsley> # ffmpeg -t 1 -f lavfi -i frei0r_src=filter_name=partik0l:filter_params=1234 /tmp/frei0r_src.webm
[23:49:37 CET] <durandal_1707> blap: how default dynaudnorm parameters cause distortions?
[23:54:22 CET] <blap> i don't know! the manual seems to indicate it never will
[23:57:56 CET] <durandal_1707> blap: with what kind of audio?
[23:58:23 CET] <blap> music, mpv
[23:58:47 CET] <durandal_1707> what kind of music?
[23:58:50 CET] <blap> seems fine with a bit lower settings though
[23:59:00 CET] <blap> orchestral, electronic
[23:59:40 CET] <blap> i spose i could test it with ffplay and mp3s
[00:00:00 CET] --- Sun Nov 12 2017