Hi there,

I'm making a basic video player in c++ using libffmpeg and SDL and I'm having some problems getting the correct Presentation Time Stamp (PTS) of each decoded frame in the stream.

I've read through the entire tutorial here: http://dranger.com/ffmpeg/ and my code is fairly similar although more object orientated.

I stream the packets into two separate queues: one for audio and one for video. A thread (the video thread) then picks up the packets in the video queue, decodes them and adds the result images to a display queue. Each of the images in the display has a value describing the number of seconds along the timeline when the frame is to be displayed (PTS). The problem I'm having is calculating this value!

For the most part the video player works and audio seems to be completely fine. here is the code in the video thread:

while( mVideoThreadRunning )
{
AVPacket packet;
if( mVideoPacketQueue.Pop( &packet ) )
{
int frameFinished = 0;
 
if( avcodec_decode_video2( mVideoCodecContext, mFrame, &frameFinished, &packet ) == -1 )
{
exit(0);
}
 
assert( packet.dts != AV_NOPTS_VALUE );
// use the packet decode timestamp as pts
double pts = (double)packet.dts;
 
// multiply by time base to get the time offset
pts *= av_q2d( mVideoCodecContext->time_base );
 
if( frameFinished )
{
// wait until the display queue frees up
while( mDisplayQueue.Size() >= 100 )
{
SDL_Delay(10);
}
 
// allocate a video frame to be displayed at 'pts' time
VideoFrame frame( mVideoCodecContext->width, mVideoCodecContext->height, pts );
 
// convert the frame data to mOutput (rgb24)
sws_scale(mSwsContext, mFrame->data, mFrame->linesize, 0, mVideoCodecContext->height, &frame.mData, &frame.mPitch);
 
cout << frame.mPts << endl;
// add the frame to the display queue
mDisplayQueue.PushSort( frame, [](const VideoFrame& a, const VideoFrame& b) -> bool
{
return a.mPts < b.mPts;
});
}
}
}
 

when I print out the pts of the first 6 frames I get this:

334

417.5

501

584.501

668.001

751.501

and the time base is 1001/48000 (AVRational) which is just under 48fps (~0.02), so for example when I calculate the pts for the first frame (334) it gives:

334 * av_q2d( time_base )[0.02] = 6.68 so the first frame only shows up after 6.68 seconds in the timeline.

This code works for roughly half the videos I've tested on it but this problem keeps creeping up. Is there any other calculations I need to perform to get the correct PTS every time? I wish there was more documentation on this.

Cheers.

Thank you so much Cornstalks!

It was because I was using the mVideoCodecContext->time_base instead of the stream :)

As for the implementation; I currently sync the video with a master clock but the audio plays out of sync from this (using SDL_AudioSpec) I'm going to change this in the future so that the audio also syncs with the master clock.

there was an audio delay on some of the videos I tested but I also wasn't talking into account the start_time you mentioned so that could have been the problem.

I've not tested with many different formats so far, mainly just mp4's

Again, thanks so much!