The timestamps on frames are applied by the kernel, so they're much more accurate than userspace would be. We typically see "a few tens of microseconds of jitter" on them, whereas in userspace you could get many milliseconds. With Linux there aren't any absolutely hard constraints that I know of, but in practice I don't see any problems here.
I think that means that an interpolation scheme should be perfectly feasible. Personally I still quite like the idea of explicitly gathering a list of all and only the timestamps that you want. You could even override the codec's encode method to (a) call the parent encode method and (b) add the wallclock timestamp of that frame to a list. But what you're proposing sounds fine to me too.
I think that means that an interpolation scheme should be perfectly feasible. Personally I still quite like the idea of explicitly gathering a list of all and only the timestamps that you want. You could even override the codec's encode method to (a) call the parent encode method and (b) add the wallclock timestamp of that frame to a list. But what you're proposing sounds fine to me too.
Statistics: Posted by therealdavidp — Thu Mar 27, 2025 12:01 pm