We currently generate mp4 fragments around flush() calls, which correspond to segment boundaries when the input in MPEG2-TS. This causes us to split GOPs, which means the later fragment is missing the reference frame and we have to drop some video data. If we fragmented on keyframes instead, we wouldn't have this problem.
We currently generate mp4 fragments around flush() calls, which correspond to segment boundaries when the input in MPEG2-TS. This causes us to split GOPs, which means the later fragment is missing the reference frame and we have to drop some video data. If we fragmented on keyframes instead, we wouldn't have this problem.