Doubts about timing and lost frame/packet at the codec level

[leandro]

As far as I understand, the h264 codec holds information about the frame rate (within the NAL, of SEI's type, timing_info_present_flag -> time_scale) and with that plus the frames count a container can build up its timing scheme. (like PTS) Is that right?

Here are some other questions:

1. Why do we have to divide the time_scale value by 2 to get the framerate?
2. Why the codec won't use the original framerate?
3. is it because we can get a double precision? (I think this is done by utilizing time_scale and num_units_in_tick)
4. how does double precision framerate are stored at the codec level? (I think this is done by utilizing time_scale and num_units_in_tick)
5. Does the codec signalizes somehow that a frame/packet was lost?
6. How an encoder identify that a frame/packet or part of it was missing?
7. How does dropped frame are handled at the codec leve?

I saw an example at Internet with what seems to be a 29.97 (time_scale/num_units_in_tick)/2 but I downloaded the big bunny (30fps) and then I extracted the .h264 and there I noticed a time_scale valued with 60 and a num_units_in_tick of 1 but the video itself is 30fps, do I really need to half its value to get the right framerate? Why spend more bits where I could store its value?

[foxyshadis]

Did you check nuit_field_based_flag? The spec is obnoxiously bad at making it clear, but the actual framerate is

time_scale / (num_units_in_tick * (nuit_field_based_flag+1))

. But that hardly matters, since the container always overrides the VUI values. Most raw streams will have a generic raw 60/1 or 25/1 and be fixed in muxing.