[Ffmpeg-devel] Timestamps on Decoding

Paul Curtis pfc
Mon Oct 24 00:45:20 CEST 2005

Ronald S. Bultje wrote:

> They are probably correct, and more than tired of having to explain it
> every time again. I explained it to you in an as-simple-as-possible
> fashion in my other email, which you appear to silently ignore.

I did read your e-mail several times ... thanks, it explained a number 
of things. In addition, I've reduced the 'mplayer' playback code to the 
basics. Luckily, I don't have to rescale to correct display versus 
stored aspect ratio.

> Richard may not be very nice to you, but he is completely right. The
> timestamps that you are using are illegal, and the fact that it works is
> most likely nothing more than pure coincidence or luck. Most likely, the
> Helix SDK does timestamp smoothing to make sure the output file is
> valid.

The Helix SDK expects a start time and end time (in milliseconds) for a 
single frame of either audio or video. When I receive a "finished" frame 
from the decoder, there is no way (as several e-mail threads atest) to 
determine the presentation timestamp. I have tried every value in all 
the structures. I have used values that were available and interpolated 
the missing values.

It *is* pure luck that the container/stream combination I'm using 
happens to present a monotonic increasing DTS for the decoded audio and 
video frames. I realize that this may not hold true for all 
container/stream combinations.

But the last question was never answered: when I receive a complete raw 
frame, how can I accurately tell what it is I have received? Queueing 
the frames (audio or video) and presenting them to the Helix encoder 
wouldn't be a problem IF I could determine when the frame SHOULD be 
presented. This is the crux of the problem, and even Richard has said 
it's a problem.


More information about the ffmpeg-devel mailing list