Recherche avancée

Sur d’autres sites (11609)

• avformat_open_input fails only with a custom IO context

  19 janvier 2017, par Tim

  Running into an odd issue with avformat_open_input, it is failing with :

  Invalid data found when processing input

  But this only happens when I attempt to read the file using a custom AVIOContext.

  My custom code is as follows (error checking omitted for clarity) :

  auto fmtCtx = avformat_alloc_context();
  
  auto ioBufferSize = 32768;
  
  auto ioBuffer = (unsigned char *)av_malloc(ioBufferSize);
  
  auto ioCtx = avio_alloc_context(ioBuffer,
  
                                  ioBufferSize,
  
                                  0,
  
                                  reinterpret_cast<void>(this),
  
                                  &amp;imageIORead,
  
                                  NULL,
  
                                  &amp;imageIOSeek));
  
  
  
  fmtCtx -> pb = ioCtx;
  
  fmtCtx -> flags |= AVFMT_FLAG_CUSTOM_IO;
  
  
  
  int err = avformat_open_input(&amp;fmtCtx, NULL, NULL, NULL);
  
  </void>

  imageIOSeek is never called, but properly handles the whence parameter including the AVSEEK_SIZE option. My file data is already loaded in memory, so imageIORead is trivial (returning 0 at EOF) :

  int imageIORead(void *opaque, uint8_t *buf, int buf_size) {
  
      Image *d = (Image *)buf;
  
      int rc = std::min(buf_size, static_cast<int>(d->data.size() - d->pos));
  
  
  
      memcpy(buf, d->data.data() + d->pos, rc);
  
      d->pos += rc;
  
      return rc;
  
  }
  
  </int>

  The data being read is loaded from a file on disk :

  /tmp/25.jpeg

  The following code is able to open and extract the image correctly :

  auto fmtCtx = avformat_alloc_context();
  
  int err = avformat_open_input(&fmtCtx, "/tmp/25.jpeg", NULL, NULL);

  The project is using a minified version of libavformat including only the formats we need. I don’t believe this is the cause of the problem since the file can be open and handled properly when the path is specified. I haven’t seen any configure options specifically targeting support for custom IO contexts.

  This is the image in question : 25.jpeg

• How to pipe live video frames from ffmpeg to PIL ?

  30 janvier 2017, par Ryan Martin

  I need to use ffmpeg/avconv to pipe jpg frames to a python PIL (Pillow) Image object, using gst as an intermediary*. I’ve been searching everywhere for this answer without much luck. I think I’m close - but I’m stuck. Using Python 2.7

  My ideal pipeline, launched from python, looks like this :

  1. ffmpeg/avconv (as h264 video)
  2. Piped ->
  3. gst-streamer (frames split into jpg)
  4. Piped ->
  5. Pil Image Object

  I have the first few steps under control as a single command that writes .jpgs to disk as furiously fast as the hardware will allow.

  That command looks something like this :

  command = [
  
          "ffmpeg",
  
          "-f video4linux2",
  
          "-r 30",
  
          "-video_size 1280x720",
  
          "-pixel_format 'uyvy422'",
  
          "-i /dev/video0",
  
          "-vf fps=30",
  
          "-f H264",
  
          "-vcodec libx264",
  
          "-preset ultrafast",
  
          "pipe:1 -",
  
          "|", # Pipe to GST
  
          "gst-launch-1.0 fdsrc !",
  
          "video/x-h264,framerate=30/1,stream-format=byte-stream !",
  
          "decodebin ! videorate ! video/x-raw,framerate=30/1 !",
  
          "videoconvert !",
  
          "jpegenc quality=55 !",
  
          "multifilesink location=" + Utils.live_sync_path + "live_%04d.jpg"
  
        ]

  This will successfully write frames to disk if ran with popen or os.system.

  But instead of writing frames to disk, I want to capture the output in my subprocess pipe and read the frames, as they are written, in a file-like buffer that can then be read by PIL.

  Something like this :

      import subprocess as sp
  
      import shlex
  
      import StringIO
  
  
  
      clean_cmd = shlex.split(" ".join(command))
  
      pipe = sp.Popen(clean_cmd, stdout = sp.PIPE, bufsize=10**8)
  
  
  
      while pipe:
  
  
  
          raw = pipe.stdout.read()
  
          buff = StringIO.StringIO()
  
          buff.write(raw)
  
          buff.seek(0)
  
  
  
          # Open or do something clever...
  
          im = Image.open(buff)
  
          im.show()
  
  
  
          pipe.flush()

  This code doesn’t work - I’m not even sure I can use "while pipe" in this way. I’m fairly new to using buffers and piping in this way.

  I’m not sure how I would know that an image has been written to the pipe or when to read the ’next’ image.

  Any help would be greatly appreciated in understanding how to read the images from a pipe rather than to disk.

  • This is ultimately a Raspberry Pi 3 pipeline and in order to increase my frame rates I can’t (A) read/write to/from disk or (B) use a frame by frame capture method - as opposed to running H246 video directly from the camera chip.

• C++ Extracting a h264 Subsequence from Byte Stream

  10 janvier 2017, par Simon

  I have a raw h.264 byte stream coming from an RTSP network camera. In order to get the byte stream, I catch the piped output from ffmpeg using popen() :

  ffmpeg -i rtsp://address -c:v copy -an -c:v copy -an -f h264 pipe:1

  At some point in time, I would like to start recording from the stream for a while (and save everything to an mp4 file). I want to achieve this without decoding the stream to an intermediate format (e.g., yuv420p) and encoding it back. As a first test, I just started writing the output buffer to disk after a couple of seconds. Then, I can encode the video again using

  ffmpeg -i cam.h264 -c:v h264 -an -f copy cam_out.mp4

  Here, ffmpeg complains that the first part of the data is corrupted (it still seems to be able to recover from this as it just throws away the corrupted parts). This of course makes sense as I simply start recording without looking for the start of frames etc.. Ideally, I would like to start and stop recording at the correct parts of the stream. I had a small glimpse on the h.264 format and the NAL units. Is there some simple way of detecting "good" positions in the stream to start recording ?