[Ffmpeg-devel] Timestamps on Decoding

[Rich Felker]

On Mon, Oct 24, 2005 at 07:03:32PM -0400, Ronald S. Bultje wrote:
> On Mon, 2005-10-24 at 18:47 -0400, Rich Felker wrote:
> > On Mon, Oct 24, 2005 at 12:30:17PM -0400, Ronald S. Bultje wrote:
> > > On Mon, 2005-10-24 at 12:32 -0400, Rich Felker wrote:
> > > > Again I have no idea what you're saying, but the only frames that can
> > > > possibly have dts==pts are B frames.
> > > 
> > > Oh! Sorry, we misunderstood each other then. I was talking about input
> > > == output as in "container given timestamp" == "output frame timestamp".
> > > I wasn't talking about DTS/PTS. For the reasons that you've mentioned
> > > earlier in this thread, I mostly try to not use DTS at all, actually.
> > 
> > What do you mean then by "container given timestamp" == "output frame
> > timestamp"? This is always true if the container has pts values,
> > unless you have a weird definition you're using...
> 
> Basically what you find in avi & co, in whatever muxing mode (sometimes
> dts, sometimes something ele which is neither pts nor dts).

AVI gives you dts, essentially. Originally it was meant to be pts,
which is the same for delay==0 streams, but then fools went and stored
out-of-order streams in avi, so that now it's necessary to interpret
AVI as having only dts whenever the codec uses B frames/delay.

Rich