Recherche avancée

Sur d’autres sites (8749)

• FFMPEG : Youtube streaming quality and speed issues

  25 février 2020, par Alex Paramonov

  I am trying to make a reliable stream from my Icecast/Shoutcast servers to Youtube live. The command that I use is :

  ffmpeg -v verbose -framerate 30 -loop 1 -i /var/image.jpg -re -i http://127.0.0.1:4700/radio -c:v libx264 -preset ultrafast -b:v 2250k -maxrate 6000k -bufsize 6000k -c:a copy -ab 128k -s 1920x1080 -framerate 30 -g 60 -keyint_min 60 -f flv  rtmp://a.rtmp.youtube.com/live2/xxx

  As you can see I am using recommended bitrate for Youtube, insert keyframes every 2 seconds and streaming at 30 frames per second.
  The stream is working but after running for some time two thing are happening :

  1. FFMPEG speed falls from 1x to something like 0.998x
  2. Youtube starts complaining that video stream speed is slow, markes the quality as bad and sometimes video starts buffering.

  Why is this happening ? CPU load is normal, connectivity is ok (the stream is running on a 1Gg/s dedicated server).

  Since in my example above I am streaming a single image as a logo of the stream I also tried to generate a short 30 seconds video with that image and broadcast that video instead of an image, but that did not help as well.

  The command I used for conversion :

  ffmpeg -framerate 30 -loop 1 -i /var/image.jpg -c:v libx264 -preset ultrafast -tune stillimage  -b:v 2250k -minrate 2250k -maxrate 6000k -bufsize 6000k -framerate 30 -g 60 -keyint_min 60 -t 30 out4.mp4

  And broadcast with

  ffmpeg -stream_loop -1 -i out4.mp4  -re -i http://127.0.0.1:4700/radio -c:v copy -c:a copy  -framerate 30 -g 60 -keyint_min 60 -f flv  rtmp://a.rtmp.youtube.com/live2/xxx

  ffmpeg version is 4.1.1

• Using FFmpeg change the video speed of different parts in python, I uses opencv and subprocess, but it's slow, is there another quicker way ?

  27 septembre 2020, par Jackal Zhao

  There is a video file, need to process like that : [0, 13, 2] means the 0th to 13th frames need 2x speed, according to the list below, change the speed of different video parts, and write the result to an output file using H264 compression.

  


  [0, 13, 2]
[13, 20, 1]
[20, 25, 2]
[25, 135, 1]
[135, 164, 2]
[164, 250, 1]
[250, 254, 2]
[254, 309, 1]
[309, 354, 2]
[354, 438, 1]
[438, 465, 2]
[465, 540, 1]
[540, 625, 2]


  


  Hopefully I wish the whole process could be done by FFmpeg and python(the only language I know).

  


  Now I implementd this way :

  


  Fitsr using subprocess.Popen(cmd, stdin=subprocess.PIPI) to create a FFmpeg thread, receiving input data form stdin, the using opencv to read the frames, and if the frame number suits, then write the frame data to subprocess PIPE.

  


  Here is the code :

  


  inputCapture = cv2.VideoCapture(inputFile)  # get the input Capture
inputFrameCount = int(inputCapture.get(cv2.CAP_PROP_FRAME_COUNT))
inputHeight = int(inputCapture.get(cv2.CAP_PROP_FRAME_HEIGHT))
inputWidth = int(inputCapture.get(cv2.CAP_PROP_FRAME_WIDTH))
inputFrameRate = int(inputCapture.get(cv2.CAP_PROP_FPS))
ffmpeg_command = 'ffmpeg -y -vsync 0 -f rawvideo -pix_fmt bgr24 -s %sx%s  -i - -an -r %s -vf "setpts=N/(%s*TB)" %s  "%s/FinalVideo.mp4"' % (inputWidth, inputHeight, inputFrameRate, inputFrameRate, ffmpegOutputOption, tempFolderPath)
FNULL = open(os.devnull, 'w')
process = subprocess.Popen(ffmpeg_command, stdin=subprocess.PIPE, stdout=FNULL, stderr=FNULL) # Open a ffmpeg thread
inputFrameNumber = 0
inputEquivalentNumber = 0
outputEquivalentNumber = 0
startTime = time.time()
for chunk in chunks: # get the [0, 13, 2] from the chunks list
    while inputFrameNumber < chunk[1]:
        successGettingFrame, inputFrameData = inputCapture.read() # read one frame from inputCapture
        inputFrameNumber += 1
        inputEquivalentNumber += (1 / chunk[2])
        print(successGettingFrame)
        if not successGettingFrame:
            break
        while inputEquivalentNumber > outputEquivalentNumber: # check if the frame should be writen to ffmpeg-input
            process.stdin.write(inputFrameData)  # write the frame data to ffmpeg stdin
            outputEquivalentNumber += 1
inputCapture.release()
cv2.destroyAllWindows()
process.stdin.close()
process.wait()



  


  But for each frame, it need to read the frame data using opencv, the write the data to ffmpeg stdin using subprocess, so the speed is really slow, it's about 0.5x speed comparing to ffmpeg libx264 compression.

  


  So I wish there could be a faster way to alter the speed of the specified parts in video.

  


  Could you guys provide some help ?

  


• FFMPEG Implement RTSP Client, high speed playback

  11 juillet 2021, par TTGroup

  I am writing software to play videos that have been recorded from NVR. I have completed most of the work, but there is one more feature that allows the user to change the play speed such as 0.5x, 2x, 4x, 8x ...

  


  I searched the internet all day and still couldn't find any suggestions. Here is my summary code below.

  


  auto pFormatCtx = avformat_alloc_context();

av_dict_set_int(&opts, "rw_timeout", 5000000, 0);
av_dict_set_int(&opts, "tcp_nodelay", 1, 0);
av_dict_set_int(&opts, "stimeout", 10000000, 0);
av_dict_set(&opts, "user_agent", "Mozilla/5.0", 0);
av_dict_set(&opts, "rtsp_transport", "tcp", 0);
av_dict_set(&opts, "rtsp_flags", "prefer_tcp", 0);
av_dict_set_int(&opts, "buffer_size", BUFSIZE, 0);

int err = avformat_open_input(&pFormatCtx, fullRtspUri, NULL, &opts);
if(err < 0)
    return;
    
err = avformat_find_stream_info(pFormatCtx, NULL);
if (err < 0)
    return;
pFormatCtx->flags |= AVFMT_FLAG_NONBLOCK;
pFormatCtx->flags |= AVFMT_FLAG_DISCARD_CORRUPT;
pFormatCtx->flags |= AVFMT_FLAG_NOBUFFER;       
av_dump_format(pFormatCtx, 0, fullRtspUri, 0);
    
int videoStreamInd = -1;
for (int i = 0; i < pFormatCtx->nb_streams; i++)
{
    AVStream* stream = pFormatCtx->streams[i];
    if (stream->codecpar->codec_type == AVMEDIA_TYPE_VIDEO)
    {
        if (videoStreamInd == -1)
        {
            videoStreamInd = i;
            break;
        }
    }   
}

if (videoStreamInd == -1)
    return; 
auto videoStream = pFormatCtx->streams[videoStreamInd];

isRunning = true;
while(isRunning)
{
    ret = av_read_frame(pFormatCtx, avPacket);
    if (ret < 0)
        return; 

    if (avPacket->stream_index != videoStreamInd)
        continue;
    
    //Code for render process here............
}


  


  I have read through this NVR API documentation and see support for 2x, 4x speed play as below

  


  Play in 2× Speed:
PLAY rtsp://10.17.133.46:554/ISAPI/streaming/tracks/101?starttime=20170313T230652Z&endtime=20170314T025706Z RTSP/1.0
CSeq:6
Authorization: Digest username="admin", 
realm="4419b66d2485", 
nonce="a0ecd9b1586ff9461f02f910035d0486", 
uri="rtsp://10.17.133.46:554/ISAPI/streaming/tracks/101?starttime=20170313T230652Z&endtime=20170314T025706Z", 
response="fb986d385a7d839052ec4f0b2b70c631"
Session:2049381566;timeout=60
Scale:2.000
User-Agent:NKPlayer-1.00.00.081112

RTSP/1.0 200 OK
CSeq: 6
Session: 2049381566
Scale: 2.000
RTP-Info: url=trackID=1;seq=1,url=trackID=2;seq=1
Date: Tue, Mar 14 2017 10:57:24 GMT


  


  How to play RTSP video with speeds of 0.5x, 2x, 4x ...?
Everyone who can assist me in this case, I am very grateful.