On Tue, 3 Mar 2015 18:01:12 +0300 Max Vlasov <max.vla...@gmail.com> wrote:
> On Tue, Mar 3, 2015 at 2:47 PM, wm4 <nfx...@googlemail.com> wrote: > > > On Tue, 3 Mar 2015 12:27:51 +0300 > > Max Vlasov <max.vla...@gmail.com> wrote: > > > > > > > > It turns out that it is just a get accessor for best_effort_timestamp > > field > > > and the corresponding set accessor (av_frame_set_best_effort_timestamp) > > is > > > called twice (in avcodec_decode_video2 and in avcodec_decode_audio4) in > > the > > > utils.c with a similar lines > > > > As I said, it's only a helper. What you really should do is always > > using the PTS, and if the PTS is not available, use the DTS. The extra > > heuristics are useful only for broken files and such; and they can be > > better than what best_effort_timestamp does. > > > > > > You probably mean pkt_pts, pkt_dts from the frame, right? At least I see > that this values used by this simple function (guess_correct_pts) as > incoming data for comparison. And also quick test shows that > best_effort_timestamp is equal to (if pkt_pts valid then pkt_pts else > pkt_dts) value most of the time. Yes. > I also see that guess_correct_pts is the mutating function relying on the > statistics of decoding which might be different for different seek/play > scenarios. > > So it looks like best_effort_timestamp not so great after all, but It's > strange that no simple function was implemented so far to implement this > simple conditional (pkt_pts first, pkt_dts next). Maybe because it's trivial. There are worse things in libav* APIs. _______________________________________________ Libav-user mailing list Libav-user@ffmpeg.org http://ffmpeg.org/mailman/listinfo/libav-user
