Recherche avancée

Sur d’autres sites (14519)

• FFmpeg encoding through h264_qsv : pts < dts

  29 décembre 2023, par Mario

  I'm modifying a portion of code that encodes input frames in an output mp4 file via the h264 software encoder which works quite well. Most of the code comes from https://ffmpeg.org/doxygen/6.0/mux_8c-example.html.

  &#xA;

  I changed the "working" code to use the h264_qsv hardware accelerated encoder. After a lot of work the new code produces h264 encoded frames. For testing purpose I can write them with

  &#xA;

  &#xA;

  fwrite(pkt->data, pkt->size, 1, fout)

  &#xA;

  &#xA;

  and mpv displays frames correctly (with some warnings).

  &#xA;

  When I use regular :

  &#xA;

  &#xA;

  av_interleaved_write_frame(fmt_ctx, pkt)

  &#xA;

  &#xA;

  (as from previously working code) returns an error and on stderr :

  &#xA;

  &#xA;

  [mp4 @ 0x7f82a80b3b40] pts (-1574122160956548608) < dts&#xA;(1574122160956547584) in stream 0

  &#xA;

  &#xA;

  To look for the cause I added the following lines just before av_interleaved_write_frame(). I read that it may be a non-fatal error and so I continue after the errors, but there isn't output.

  &#xA;

  &#xA;

  frame->pts=210
  &#xA;pkt->pts=-1574122160956548608
  &#xA;pkt->dts=1574122160956547584
  &#xA;[mp4 @ 0x7effb40b3b40] pts (-1574122160956548608) < dts (1574122160956547584) in stream 0
  &#xA;...
  &#xA;frame->pts=240
  &#xA;pkt->pts=-1574122160956548608
  &#xA;pkt->dts=1574122160956548096
  &#xA;[mp4 @ 0x7effb40b3b40] pts (-1574122160956548608) < dts (1574122160956548096) in stream 0

  &#xA;

  &#xA;

  I did another test assigning incremental values to pkt->pts and pkt->dts, it works but resulting fps is wrong, and I don't think it is the right solution !

  &#xA;

  Why does the av_interleaved_write_frame(fmt_ctx, pkt) function returns an error with h264_qsv encoder ? Who should set pkt->pts and pkt->dts ?

  &#xA;

• FFMpeg C++ How to create a filter with multiple outputs ?

  11 septembre 2020, par GoodSimon

  For example, we have one AVFrame with a size of 128x128 pixels and the task is to split the AVFrame into 4 parts (4 separate AVFrame). To do this, I fill the graph with filters using the avfilter_graph_parse2(...) function, and then call avfilter_graph_config(...).

  &#xA;

  Let's start. Let's cut out the top left corner. To crop a frame, we need to use the crop filter, which means we initialize the AVFilterGraph graph with the line :

  &#xA;

  buffer = width = 128: height = 128: pix_fmt = 2: time_base = 1/25, pixel_aspect = 128/64 [pad_in];
[pad_in] crop = w = 64: h = 64: x = 0: y = 0, buffersink;


  &#xA;

  Everything works great ! Now let's try to make several outputs :

  &#xA;

  buffer = width = 128: height = 128: pix_fmt = 2: time_base = 1/25, pixel_aspect = 128/64 [pad_in];
[pad_in] crop = w = 64: h = 64: x = 0: y = 0, buffersink;
[pad_in] crop = w = 64: h = 64: x = 64: y = 0, buffersink;
[pad_in] crop = w = 64: h = 64: x = 0: y = 64, buffersink;
[pad_in] crop = w = 64: h = 64: x = 64: y = 64, buffersink


  &#xA;

  As you can see, we have one buffer for the input image, 4 crop filters for cutting each piece, and 4 buffersink filters for the output images. The call avfilter_graph_parse2(...) returns 0, which is good, but avfilter_graph_config() returns the error code -22 == AVERROR(EINVAL) and the message is output to the console : Input pad "default" with type video of the filter instance "Parsed_crop_3" of crop not connected to any source.

  &#xA;

  I am asking for your help in creating a filter with multiple outputs.

  &#xA;

• Camera's pts and dts jumps issue

  14 mars 2014, par tilzaer

  I use ffmpeg library in my code for capturing rtsp streams from cameras and writing in flv. If I capture streams from only one camera and camera has only video stream then there's no any errors, I got first packet with pts and dts 1698557894 and other packets' pts and dts slowly increases. But if camera has video and audio streams, then strange things occur. For example, video packets pts and dts begin with 1698557894 and slowly increase, and audio starts with 0 then slowly increases and after 50 packets jumps to value 151004317 and slowly increase. Another situation, when audio begins with 0 and slowly increases, and video begins from 1785662594 and after 70 packets jumps to 234722 and slowly increase. Such behaviour do not allow flv segment muxer to write files, it just returns some EINVAL value.

  Also if I try to capture streams from two cameras, first camera has only video and another has video and audio, then first camera packets' pts and dts are ok. But another camera has video and audio pts/dts values which are very different. av_read_frame returns video packet with pts/dts value 1811924055, which rescales to 557003451 in flv muxer and audio 4456027604, which rescales to 557003451, but these values must be almost equal !

  So, the questions are :
  0) why do these jumps occur in the beginning of capturing ? is it a problem with camera or it is just some ffmpeg issue ?
  1) can such jump occur after some long period ? how I should handle it ?
  2) why camera's video and audio have so different pts/dts values ?