Recherche avancée

Sur d’autres sites (7828)

• FFmpeg muxing theora/vorbis unable to flush ?

  11 novembre 2013, par user2979732

  I'm pretty new to ffmpeg and it's confusing. I'm working on a basic muxer and have been spending over a week on this - I don't normally post as I solve 98% of my issues with google, but unable to get this one so far.

  The basis of my source is FFmpeg's own muxing.c example. When I try to force it using libvorbis for audio, and create "test.ogg" it demonstrates the same issues I'm having in my own derivation of muxing.c. The problem is with ogg/theora/vorbis. I'm forcing the use of audio codec like this :

  audio_st = add_stream(oc, &audio_codec, avcodec_find_encoder_by_name("libvorbis")->id);

  It seems the problem is in not setting audio pts in the muxing.c sample. There is a confusion in general about this, nobody apart from this guy didn't address what I am looking for http://webcache.googleusercontent.com/search?q=cache:6ml82RMN3YYJ:ffmpeg.org/pipermail/libav-user/2013-April/004304.html+&cd=4&hl=en&ct=clnk&gl=cz

  I couldn't find any answers to that naturally - like why don't they set the audio pts ? Laziness ? Not needed ? Do they believe all encoders will produce the pts for them(not true as seen below) ?

  Anyway, when you try muxing.c with mp4/libx264/forced libmp3lame all is fine, but the encoder says that "encoder did not produce valid pts, making some up.". However, it's silent with ogg/theora/vorbis, as if there were valid pts(?) but the result is no audio packets present in the stream(!), at least from what I saw using ffprobe. Which results in the video not being able to replay even, until you take out the empty audio stream. Then it plays the video, which shows that stream is fine.

  Coming to my original issue. I tried setting the pts on the audio frame you're sending to the encoder to fix that problem(this already sucks). I was unable to find a definite answer how to properly set the pts - that's the other big issue as I'm trying stuff which I'm not sure works. Anyway, in the end when setting "some" pts, this results in ogg with sound.

  if (frame->pts == AV_NOPTS_VALUE) frame->pts = audio_sync_opts;
  
  audio_sync_opts = frame->pts + frame->nb_samples;

  I'm aware I should probably use rescaling to adjust for the container time bases etc..if this was present/explained in ffmpeg's own sample I wouldn't have to guess now (as I'm stil not 100% sure about time base relationship between container and codec, I think container time base takes somehow over the codec one).

  My other problem is flushing - but that might have something to do with the screwed up pts. So I won't rather get into that in detail - the basic problem is, when I send finite number of audio frames, like 20, I get 2 packets only for example. From my understanding, I need to flush the rest of audio after all the encoding/muxing is done, which I managed to do with mp4/libx264/libmp3lame, but with ogg/theora/vorbis it doesn't flush. Why not, I have no idea.

  If someone could rework muxing.c into sending it finite number of audio / video frames - ie . not until duration > X, but until it sent 20 video & 100 audio frames(just an example). So that number of frames I have is important, not the video time I end up with. Then encode / mux all the frames - with proper video/audio pts, working with theora/ogg and flushing if needed, that would probably solve all of my issues. I'm sure for an expert ffmpeg'er modifying muxing.c addressing all those things would be a pretty quick exercise and could help more than 1 confused person.

  Thanks !

• Live video encoding using...?

  27 décembre 2013, par Basic

  I'm attempting to write a fairly simplistic application that will stream video/audio from a webcam to someone else across the internet (ala Skypebut with more control).

  There seems to be very little useful/relevant information on the subjectand what I can find is largely outdated. From my research so far, x264 seems to be the way to go as it offers an ultrafast option which is designed for this situation

  I'm able to turn on the webcam and receive a stream of images. I can also listen on an audio device and get samples.

  Where I'm failing is encoding that information in such a way as to be able to stream with a minimum of latency (from what I've read, 200ms delay is the goal for no obvious lag, including network latency - so let's aim for 100-150ms)

  Things I've tried

  ffmpeg

  This seems to be the most widely used option for encoding. I've had two real issues using it. Firstly, even using x264 with no look-aheads and the bare minimum buffers for stability, the delay seems to be on the order of 700ms using image2pipe. Secondly, it requires ffmpeg to be installed - being able to do this without an external dependency would be nice.

  VLC

  As with ffmpeg this requires an external program which is a negative. Even worse, I can't seem to get a latency of under 2 seconds which seems to increase over time. I've also only been able to get VLC to capture the camera itself rather than take a stream of images which means I don't get a chance to pre-process them.

  DirectShow

  I've seen a number of sites recommending using the windows direct show encoders but I haven't been able to find one that works at anything like real time. In fact, the only one I've managed to get going reliably is a Windows Media codec that has a massive latency and fairly large size.

  Other considerations

  None of the above address the problem of adding an audio stream to the video. I'm not sure if I should attempt to encode them together or send a separate stream alongside the video.

  In short, I've been Googling for a week or so now and haven't found a decent way to do this. Can someone please point me at a decent example/guide ?

• OpenCV 2.4.7 with FFMpeg support build with VS 2010 (x86)

  24 novembre 2013, par GregPhil

  I'm fighting with new version of OpenCV 2.4.7 for almost one week (prev. used 2.4.2). My motivation for the new setup is the fact that I run into problems using the Videowriter for uncompressed grayscale videos (http://answers.opencv.org/question/3603/videowriter-videocapture-and-uncompressed-avi/). In https://github.com/Itseez/opencv/pull/1516 "akarsakov" is recommending to build OpenCV without vfw support and open uncompressed video through ffmpeg. That's what I tried to do. I followed basically the instructions mentioned in this video : http://docs.opencv.org/trunk/doc/tutorials/introduction/windows_install/windows_install.html

  I had successfully build opencv with Qt 5.1.1 (http://answers.opencv.org/question/24184/how-to-compile-opencv-247-with-qt-and-ffmpeg/). But I'm not able to get the VideoWriter & VideoCapture work for me. So what I did :

  1. I downloaded the newest ffmpeg build from http://ffmpeg.zeranoe.com/builds/

  2. I generated the vs2010 project disabling VfW and enabling ffmpeg instead :

     Video I/O:
     
     
     
     Video for Windows:           NO
     
     DC1394 1.x:                  NO
     
     DC1394 2.x:                  NO
     
     FFMPEG:                      YES (prebuilt binaries)
     
       codec:                     YES (ver 55.18.102)
     
       format:                    YES (ver 55.12.100)
     
       util:                      YES (ver 52.38.100)
     
       swscale:                   YES (ver 2.3.100)
     
       gentoo-style:              YES
     
       OpenNI:                    NO
     
     OpenNI PrimeSensor Modules:  NO
     
     PvAPI:                       NO
     
     GigEVisionSDK:               NO
     
     DirectShow:                  YES
     
     Media Foundation:            NO
     
     XIMEA:                       NO

  3. I added the include directories and libs for ffmpeg to the highgui module.

  I was not able to compile OpenCV without any errors but the behaviour what I see is not what I expect :

  1. Build OpenCV 2.4.7 without VfW -> VideoWriter.open () returns always false !
  2. Build OpenCV 2.4.7 with VfW -> VideoWriter.open () returns true for the time, calling a release() and an open() on the same object returns always a false.
  3. With or without ffmpeg -> An VideoCapture.open () gives a error message asking for libiconv-2dll !

  Does some had the same problems and is able to point me to my mistake ?

  Thank you in advance.
  cheers