I need to remux a live stream from an RTSP container to framgented MP4.
Video stream is a h264 baseline one.
I can have a very good results with
ffmpeg, however, I'm struggling with one key thing:
I would like to have a very quick first frame (ie. very short time-to-first-frame).
Now my video source encoder puts an IDR-frame into the stream at every 10 seconds. This one we can consider as a keyframe. I also modded the encoder to immediately put an IDR frame into the stream, once a client joins to it over rtsp.
I applied these flags to ffmpeg:
+frag_keyframe+empty_moov+default_base_moof, and created a pipeline:
CAM -> RTSP(H264) -> FFMPEG -> fMP4(H264) -> VLC player
Due to above config, I will get frames written at every keyframe, and for somewhat reason the first keyframe is not taken into account, so whenever I start the process, VLC will be silent for 10seconds (then comes the next keyframe, and buffers are flushed), VLC will show a still picture, another 10s needs to be passed (then comes again the next keyframe), and after that VLC just plays fine with a 20s lag.
I can use
frag_every_frame, which works just perfect, but it has a massive performance drawback.
I tried to use
frag_custom, and flush the buffers (creating fragments) at 1Hz in the beginning for fast startup, then slow down the frequency to fragment at every 10s: but in this case, fMP4 player first starts correctly, but after the change of fragment timings, it stalls.
How can I change fragmentation on the fly in case a fragmented MP4 stream?