[Ffmpeg-devel-irc] ffmpeg.log.20160722

burek burek021 at gmail.com
Sat Jul 23 03:05:01 EEST 2016


[00:07:04 CEST] <DelphiWorld> still zero luck
[00:07:48 CEST] <DelphiWorld> even if i do format conversion in the vaapi filtergraph it's not applied
[00:19:30 CEST] <DelphiWorld> furq: finaly that cmd work with both flv & mp4 but dont want to stream to rtmp. strange
[00:19:38 CEST] <DelphiWorld> ffmpeg -vaapi_device /dev/dri/renderD128 -i udp://@239.100.1.2:1234 -acodec libfdk_aac -profile:a aac_he_v2 -b:a 64k -ar 48k -pix_fmt yuv420p -vf 'format=nv12|vaapi,hwupload,scale_vaapi=640:360' -c:v h264_vaapi -vb 800k -f flv rtmp://cdn.mybox.net/live/almajd
[00:24:19 CEST] <furq> can you stream anything to that rtmp server
[00:24:32 CEST] <DelphiWorld> furq: let me try to stream a video with copy
[00:26:20 CEST] <DelphiWorld> yes furq it's streaming
[00:30:16 CEST] <furq> encode an flv with those settings and pastebin the output of ffprobe
[00:30:24 CEST] <DelphiWorld> ok
[00:37:53 CEST] <DelphiWorld> i even lose the settings... lot of try's
[00:50:44 CEST] <DelphiWorld> furq: are we missing hwdownload after the scale filter?
[00:51:21 CEST] <furq> you can try it but i got the impression that was for using a software encoder
[00:51:35 CEST] <DelphiWorld> ok so no need it
[00:51:39 CEST] <DelphiWorld> i want fully hardware
[00:51:58 CEST] <furq> i don't have a vaapi device to test with so i'm just reading the same docs as you
[00:52:09 CEST] <DelphiWorld> furq: i can provide a box
[00:52:42 CEST] Action: DelphiWorld is encoding the flv
[00:52:57 CEST] <DelphiWorld> i even removed the scalling...
[00:53:00 CEST] <DelphiWorld> ffmpeg -vaapi_device /dev/dri/renderD128 -i udp://@239.100.1.2:1234 -acodec libfdk_aac -profile:a aac_he_v2 -b:a 64k -ar 48k -vf 'format=nv12|vaapi,hwupload' -c:v h264_vaapi -vb 800k  -f flv out.flv
[00:54:10 CEST] <DelphiWorld> furq: http://paste.ubuntu.com/20374150/
[00:55:02 CEST] <jkqxz> Could the default 2 B-frames be bad for streaming?  The reordering might cause problems somewhere, though I don't know what.  (That is, try adding "-bf 0".)
[00:55:38 CEST] <DelphiWorld> jkqxz: ok, where to add this -bf 0? after or befaure the -codec:v ?
[00:56:00 CEST] <jkqxz> After -c:v, next to the bitrate.
[00:57:28 CEST] <DelphiWorld> jkqxz: zero effect..., still shutting down
[00:57:33 CEST] <DelphiWorld> ffmpeg -vaapi_device /dev/dri/renderD128 -i udp://@239.100.1.2:1234 -acodec libfdk_aac -profile:a aac_he_v2 -b:a 64k -ar 48k -vf 'format=nv12|vaapi,hwupload' -c:v h264_vaapi -vb 800k -bf 0 -f flv rtmp://cdn.mybox.net/live/almajd03
[00:58:24 CEST] <DelphiWorld> thought it work when multicasted, but rtmp refuse it
[00:58:28 CEST] <jkqxz> Does it work with a software encode (libx264)?
[00:58:38 CEST] <DelphiWorld> yes it work
[00:58:44 CEST] <furq> bframes should make no difference with rtmp
[00:59:13 CEST] <furq> i have no idea why outputting to an flv file would work, but rtmp wouldn't
[00:59:57 CEST] <DelphiWorld> gonna make a flv based on that then try to rtmp it and see
[01:00:21 CEST] <furq> if that doesn't work then i'd have to suspect there's a bug somewhere
[01:00:27 CEST] <furq> er
[01:00:29 CEST] <furq> if that does work, rather
[01:01:52 CEST] <DelphiWorld> http://paste.ubuntu.com/20374905/
[01:01:53 CEST] <DelphiWorld> dont work
[01:05:13 CEST] <DelphiWorld> i used -vn, audio works
[01:05:17 CEST] <DelphiWorld> but -an, video dont work
[04:22:47 CEST] <GyrosGeier> hi
[04:23:05 CEST] <GyrosGeier> I'm trying to convert a capture from my laptop camera to mpeg4
[04:23:24 CEST] <GyrosGeier> it is horribly slow for some reason, something like 0.1fps
[04:24:21 CEST] <GyrosGeier> encoding speed is normally around 10-20fps for me
[04:24:31 CEST] <GyrosGeier> with 720p content
[04:24:42 CEST] <GyrosGeier> now I have 640x480
[04:24:58 CEST] <GyrosGeier>   Stream #0:0 -> #0:0 (rawvideo (native) -> mpeg4 (native))
[07:47:52 CEST] <mrelcee> trying to compile 3.1.1 freebsd 10.3 anyone care to help?
[07:48:01 CEST] <mrelcee> +on
[08:22:46 CEST] <relaxed> mrelcee: what's the problem?
[08:24:44 CEST] <mrelcee> relaxed:  hi thanks..    i successfully .configure. with my options   but when i type make   i get a long list of "line X: Need an Operator".      and a fail
[08:26:55 CEST] <mrelcee> http://pastebin.com/k1V4YPzv
[08:27:58 CEST] <mrelcee> i installed the latest make after the first fail.  changed nothing.  I suspect I didn't upgrade anything there.
[08:28:09 CEST] <relaxed> mrelcee: freebsd's `make` is not gnu make- use gmake
[08:28:31 CEST] <mrelcee> ooo.  it's compiling..
[08:30:12 CEST] <mrelcee> thanks for the help..   hopefully it all continues smoothly
[08:35:16 CEST] <relaxed> you're welcome
[08:37:58 CEST] <mrelcee> you're my new favorite person.  :)
[08:38:20 CEST] <mrelcee> i've been beaitng my head on this for a few hours..
[08:45:16 CEST] <mrelcee> It successfully compiled.   yay.   thanks again..
[11:00:12 CEST] <wjvds> I am trying to stream from v4l2 to ffserver but when I use the server (http://localhost:8090/feed1.ffm) as output I get the error  Input picture width (1920) is greater than stride (736) Video encoding failed but when I output to a file I (test.mp4) everything works fine
[11:02:44 CEST] <wjvds> My full command and output is http://pastebin.com/eQneFz3G
[11:02:57 CEST] <mgbowman> is ffserver still maintained?
[11:41:49 CEST] <platzhirsch> JEEB: asked on the mailing list and I guess you're right but speed could be utilized for monitoring purposes if I understood it now correctly, because if it falls below 1.0 the transcoding cannot keep up (in terms of live streaming)
[11:45:30 CEST] <JEEB> yeah I was mostly noting that the quantizer value shouldn't be used
[11:45:39 CEST] <JEEB> the rest could be utilized I guess
[14:58:06 CEST] <osiris> hi guys
[14:58:57 CEST] <osiris> honestly i am googling for ours already, is there a place, where I can download compiled ffmpeg version including vvapi and/or intel quick sync?
[15:04:30 CEST] <jkqxz> Distributions may make builds with vaapi.  A prebuilt version outside distribution setup doesn't really work because libva is really just a dlopen() wrapper.
[15:07:09 CEST] <osiris> jkqxz: isnt it a intel based thing? I looked into the building process and its reall huge effort, and if static versions are delivered with some media servers, i thought an ffmpeg with quick snyc or something should be availeable somewhere as well
[15:07:13 CEST] <jkqxz> The Intel stuff is non-free, so you can't distribute a binary linked with libmfx.  Linking with the free mfx_dispatch can work, but you need a matching media sdk install to use it so it wouldn't really be very useful.
[15:07:46 CEST] <osiris> ah, that explains a lost
[15:08:24 CEST] <osiris> intel should cahnge their mind and give a better license modell, that sucks and they would benift from it
[15:08:38 CEST] <osiris> s/benift/benefit
[15:10:07 CEST] <jkqxz> The open source stuff with vaapi is all fine licence-wise, it's just that the driver dependency messes up making a static build.  Does your distribution not contain a suitable version already?
[15:11:01 CEST] <osiris> my media server is 15.10 ubuntu
[15:11:08 CEST] <osiris> but will be 16.04 soon
[15:11:12 CEST] <osiris> or maybe arch
[15:11:41 CEST] <osiris> butt vaapi isnt built in the ffmpeg of ubuntu
[15:11:46 CEST] <osiris> as far as i can check
[15:12:35 CEST] <osiris> no --enable=vaapi
[15:14:21 CEST] <osiris> and i have vaapi active with kodi already, but try to get transcoding working with ffmpeg now
[15:16:10 CEST] <osiris> so do u know how I get ffmpeg to make use out of it?
[15:16:16 CEST] <osiris> vainfo: VA-API version: 0.39 (libva 1.7.1)
[15:17:02 CEST] <jkqxz> Hmm.  Ubuntu 16.04 is still on the 2.8 branch and current development is on the 3.0 branch, so that doesn't help.  (The vaapi support in the ffmpeg utility is recent; added between 3.0 and 3.1.)
[15:17:35 CEST] <jkqxz> So I think you'd need to build it yourself to use on those distributions.
[15:17:53 CEST] <osiris> damn
[15:18:26 CEST] <osiris> usually there is always a "special" ppa
[15:19:54 CEST] <jkqxz> There probably is somewhere, but nothing official as far as I know.
[15:19:56 CEST] <osiris> but thx for answering, really apreciate it
[15:20:21 CEST] <osiris> dont need official ;)
[15:52:24 CEST] <osiris> this library installing is awfull
[15:52:46 CEST] <osiris> feels like already 100 installed and still some missing
[16:38:25 CEST] <soulshock> anyone played with Netflix's VMAF? I'm wondering if there's a way to avoid creating YUV files for each comparison and maybe generating them on-the-fly
[17:41:43 CEST] <platzhirsch> xit
[18:37:07 CEST] <whald> can someone tell me if a h264 frame that was handed to me through avcodec_receive_frame may still be read from the h264 decoder?
[18:37:34 CEST] <whald> (i assume that it will at least not write to the buffers after giving me the frame)
[18:40:55 CEST] <jkqxz> Yes.  The H.264 decoder will keep a reference to it and read from it for as long as it has to be kept to decode other frames (which can be arbitrarily long).
[18:43:02 CEST] <whald> jkqxz, i kind of expected that. thanks for the clarification.
[18:44:16 CEST] <jkqxz> If you want to write on a frame from the decoder, av_frame_is_writable() can test whether you hold the only reference and av_frame_make_writable() will make a copy if necessary to leave you with a writable frame.
[18:45:37 CEST] <whald> jkqxz, i'm decoding diretly to dma-bufs allocated by an intel gpu and think i have some cache cohereny problems causing artifacts on the screen. but the performance is sweet. :-)
[18:47:31 CEST] <whald> jkqxz, i'm doing this exercise to get rid of unneccesary copies, so i'll have to find a way to do this another way. i this flushing the cpu caches or the like after avcodec_receive_frame but before telling the gpu to use the buffers might be enough
[18:47:34 CEST] <jkqxz> Yeah, ok - that only works if your frames are in normal memory.
[18:48:10 CEST] <jkqxz> You know that Intel GPUs contain a hardware decoder for H.264, which decodes directly into GPU surfaces?
[18:49:02 CEST] <whald> jkqxz, i'm using this for yuv420p, but intel does not support yuv444p, so...
[18:50:07 CEST] <jkqxz> Yeah, noone makes hardware for 4:4:4 decode.
[19:25:31 CEST] <dustobub> Hey! I'm encoding an h264 mp4 and am having a small issue with the output byte size of the first frame vs. all other frames. The first frame always ends up being ~2x the byte size of other frames. Is there any way to control this behavior? My usecase requires that the first frame be very similar in size compared to other frame. Thanks.
[19:26:28 CEST] <dustobub> ffmpeg -framerate 60 -i "C:\XYZ%4d.tif" -c:v libx264 -g 1 -vf "scale=3840:2160" -crf 19 -pix_fmt yuv420p C:\XYZ.mp4
[19:27:09 CEST] <dustobub> Using ffprobe I see that pkt_size of the first frame is always ~2x compared to other frames
[19:50:55 CEST] <DHE> dustobub: key frames are bigger than other frames. this is quite normal.
[19:51:56 CEST] <DHE> there is an option, intra-refersh, which will reduce keyframe size but it has other drawbacks
[19:52:05 CEST] <DHE> oh, -g 1. didn't see that
[19:54:28 CEST] <dustobub> DHE: but all of my frames are keyframes. Just the first frame always ends up being larger for some reason.
[19:54:55 CEST] <dustobub> If I encode a dummy black frame as the first frame I see these results:
[19:54:57 CEST] <dustobub> frame|pkt_size=2173|pict_type=I
[19:54:57 CEST] <dustobub> frame|pkt_size=466255|pict_type=I
[19:54:57 CEST] <dustobub> frame|pkt_size=430179|pict_type=I
[19:54:57 CEST] <dustobub> frame|pkt_size=416652|pict_type=I
[19:54:57 CEST] <dustobub> frame|pkt_size=411401|pict_type=I
[19:54:58 CEST] <dustobub> frame|pkt_size=407174|pict_type=I
[19:55:00 CEST] <dustobub> frame|pkt_size=405377|pict_type=I
[19:55:02 CEST] <dustobub> frame|pkt_size=403207|pict_type=I
[19:55:04 CEST] <dustobub> frame|pkt_size=401588|pict_type=I
[19:55:06 CEST] <dustobub> frame|pkt_size=401200|pict_type=I
[19:55:30 CEST] <dustobub> But without that dummy frame I see this:
[19:55:36 CEST] <dustobub> frame|pkt_size=1377043|pict_type=I
[19:55:36 CEST] <dustobub> frame|pkt_size=406953|pict_type=I
[19:55:36 CEST] <dustobub> frame|pkt_size=407200|pict_type=I
[19:55:36 CEST] <dustobub> frame|pkt_size=406647|pict_type=I
[19:55:48 CEST] <dustobub> [...]
[20:01:42 CEST] <dustobub> I posted some more details on stackoverflow - http://stackoverflow.com/questions/38532700/ffmpeg-first-frame-byte-size-over-2x-size-compared-to-all-other-frames-w-100
[20:01:59 CEST] <dustobub> But let me know if you have any ideas. Thanks.
[20:15:22 CEST] <apoms> Hi all. Is it possible to setup and initialize an AVFormatContext without probing the input file? I would like to be able to preprocess a file once to discover its format and then each time I open it in the future I could just start decoding frames without having to probe for the container and stream information.
[20:50:00 CEST] <CFS-MP3> Slightly offtopic... trying to get Intel QVC to work... I compiled ffmpeg OK, but it seems to require a linux kernel patch that might not be available for 4.4... any suggestion?
[20:56:27 CEST] <jkqxz> CFS-MP3:  Change your OS to the one supported CentOS version.  (The Intel Media SDK is an immense pain on Linux.  Unless you really need the stuff in it, use VAAPI to access your QuickSync hardware instead.)
[21:04:36 CEST] <soulshock> hm what's the difference between -psnr and -filter_complex "psnr" ?
[21:05:22 CEST] <soulshock> as in what does -psnr use as reference video and what does the filter use as reference video?
[21:09:49 CEST] <CFS-MP3> jkqxz I was hoping for a solution that didn't imply switching to CentOS...that has too many implications at this point, so I think I'm going to pass until intel releases a decent patch
[21:15:18 CEST] <kepstin> soulshock: the -psnr option to the x264 encoder causes x264 to print out the psnr between the input and output of the encoder.
[21:15:40 CEST] <kepstin> i think
[21:15:45 CEST] Action: kepstin should double-check that
[21:16:55 CEST] <kepstin> might work with some other encoders too, not sure what the full list is
[21:18:55 CEST] <soulshock> x264 is fine that's what I'm using. I'm asking because I'm getting wildly different values between -psnr and -filter_complex "psnr"
[21:23:20 CEST] <soulshock> I guess what you're saying is the -psnr option only compares the frames sent to x264 whereas -filter_complex compares the input video to the output video. meaning -psnr does not take into consideration any resizing or deinterlacing
[21:24:14 CEST] <soulshock> and -filter_complex "psnr" does include any additional filters applied to the source, i.e. deinterlacing and resizing (I'm upscaling to source resolution when calculating psnr)
[21:37:03 CEST] <soulshock> now I'm thinking if it's possible to use filter "split" on output of x264 and write one to file and the other as input for psnr
[21:41:42 CEST] <DHE> soulshock: filter_complex analyzes the 2 videos using ffmpeg filters. -psnr asks x264 for its opinion on the subject
[21:42:26 CEST] <soulshock> ok. so I need filter_complex for what I want. thanks
[21:43:39 CEST] <soulshock> hm maybe not. I'm confusing myself
[21:44:22 CEST] <dustobub> per my CRF issue above I have found that the first frame when encoding with libx264 uses a QP value half of where it should be
[21:44:24 CEST] <dustobub> [libx264 @ 05380b60] frame=   0 QP=14.92 NAL=3 Slice:I Poc:0   I:32400 P:0    SKIP:0    size=1485053 bytes
[21:44:24 CEST] <dustobub> [libx264 @ 05380b60] frame=   1 QP=29.48 NAL=3 Slice:I Poc:0   I:32400 P:0    SKIP:0    size=361196 bytes
[21:44:24 CEST] <dustobub> [libx264 @ 05380b60] frame=   2 QP=29.48 NAL=3 Slice:I Poc:0   I:32400 P:0    SKIP:0    size=359406 bytes
[21:44:34 CEST] <dustobub> As discussed in http://stackoverflow.com/questions/38532700/ffmpeg-first-frame-byte-size-over-2x-size-compared-to-all-other-frames-w-100
[21:45:32 CEST] <dustobub> Does anyone know if it's possible to get CRF encoding to handle that first frame like other frames? Most people probably wouldn't care, but my use-case depends on having that first frame similar to other frames.
[21:55:55 CEST] <audioblazer> Hello, i'm new to ffmpeg and am having a weird issue when accessing a input file I use in my command after I run the line
[21:57:09 CEST] <audioblazer> It says there was invalid data found when processing the input (although was valid prior to the run) and and says moov atom not found
[22:25:49 CEST] <Mavrik> audioblazer, you have a broken video file.
[00:00:00 CEST] --- Sat Jul 23 2016


More information about the Ffmpeg-devel-irc mailing list